Techstrong TV August 29, 2025

Today's episode revolves around two letters. Can you guess them? You're watching Textron Gang, Happy Friday and Labor Day Weekend from the Gang.

Yes, we are here to talk about ai. Of course, its impact on jobs it spending, and what Silicon Valley is doing to back it politically in the upcoming 2026 midterm elections. But most importantly, we wanna wish you and yours a safe and happy Labor Day weekend as we head into it.

I'm John Swartz. I'm in Silicon Valley. I'm our senior content writer at techron.

I'm sitting in for our co-host Alan Shimmel and Mike Baard. And with me today are some of our best pundits. They're all great guy Courier, Kate Scarcella and Jack Gold.

And, uh, we're gonna start off first with something that Alan wrote. It was kind of a, uh, labor Day, or as Kay would say, unor, I think Jack said that actually Unlabored day message about AI in the state of AI and its impact on jobs. And, uh, he said this in this commentary he wrote for Tech ai, and he talked about layoffs continuing to loo large, in part because of the specter of AI replacing human roles.

And, uh, he cited a couple of statistics. All Dary, by the way, there's one tracker that found like 140,000 people. Literally 140,000 people had been impacted by nearly 500 layoffs this year.

Um, more specifically, software engineering postings, job postings have dropped significantly, no matter how you look at it. Um, any timeframe. AI related job posts have plunged and anecdotally and even proven by, by, by data.

There has been some numbers out about the college graduates in computer science and computer engineering, and a significantly higher unemployment rate or out of work rate among them. And we can get into that later. But first, uh, kay, you have some strong feelings about the industrial revolution and the, the, the phase that we're in right now.

Maybe you could fill in what you, what you were talking about earlier. Yeah, absolutely. Thank you for that, John.

So first it's important to acknowledge the, the numbers. The numbers are real, you know, and for many of us who have been in the IT field, it's, it's, um, concerning without a doubt, but it's, it's important to understand. So I, I do, I talk a lot about the industrial revolution.

Industrial Revolution. 0 was, was literally the steam engine and the replacement of humans. 0 like that.

0 and that started and was kicked off in 2020. And literally, it's about human machine teaming. It is about, it is about augmenting humans.

It is so important that we embrace, and it's something that Alan put at the end of the article about embracing, we do we need to embrace, um, ai. However, as we embrace AI, and, um, the, I think it's like the next segment or a segment after one of it talks about this doomsday. Yes.

Is, is all this real? Yes. Are the numbers real?

Is the potential of AI being dangerous? Re real? Yes.

But it doesn't have to be. We understand more now than we have because of history, because of how we have seen the, the industrial revolutions play out. And while this is new, we absolutely need to, as we move forward, just secure, embrace insecure.

As we, as we look at this. I mean, it's, it's important. It's important that we embrace ai.

It's important that we under, because humanly it is not possible for us as a person who has been in cybersecurity now for, you know, for 20 years, looking at all those network logs, looking at all the different events that was coming into the soc to the security operation centers. It was not humanly possible for us to examine all these logs and not chase false positives. And one of the things that happened as we brought in these junior analysts, and we talked about this wild, exciting career in cybersecurity, and they would chase one false positive after another.

They'd be like, really, this is what I signed up for. So, you know, this is what we need AI for. And, and I'm just talking now, but hey, guy, Jack, what do you guys think?

Yeah, I, look, I, I, I think the, the impact of AI is real. I think in some cases it's been way overestimated. I, I, I'm not convinced that AI is gonna replace all our jobs in the next few years.

Um, and I'm not convinced that it's even going to replace a lot of the entry level jobs. I think some of what we've seen, uh, the effect of, of job loss has to do with the economy or the perceived belief that the economy is not gonna be great, uh, going forward. 0.

Everyone said, no one, you know, with all the steam engines, we're not gonna need labor anymore. We're not gonna need people anymore. Guess what?

That didn't happen. 0. We change, we, we adapt.

And I think that's going to happen. Yes, there are a lot of low level entry level jobs that AI can do that can't be created. AI does not create things.

It, it, it regurgitates. It's really good at that. Uh, and so we need to be smart about how we define entry level jobs and better deal with how we educate people coming into the workforce so that they can be augmented by ai, not replaced by it.

And I think that's a real process that we haven't started yet. Or, or we haven't started very well yet, at least. Um, now that's number one.

Number two is if AI takes all the entry level jobs, that we are not hiring entry level people anymore, how do we get mid-level and high level people? They come up through the ranks. AI's not gonna do that for us.

So there are a lot of issues around this. 5. I don't know what the number would be, but we need to think about what all of this means long term.

And in fact, I think ultimately AI creates more jobs than it displaces, just as all technologies have in the past. So it, it's, Can I ask you a quick question, Jack, about that? Like, what, what do you think the timeframe is for that?

When do we see this transition from maybe initial job losses to, to job gains eventually? I think we're gonna see that in the next one to two years. I think we're already starting to see it in certain industries.

I mean, think of all, think of all the jobs that AI is created in at, at Nvidia as, as an example, right? It's, it's, you know, it's, we're, we're, or, or at AWS or, or, you know, Google or whoever in, you know, in the cloud space. Um, so think about the high level programming that gets done in ai that's not being done by AI itself, in most cases.

It's being, it's doing some low level stuff, but there's always software engineers that are looking over the shoulder and finding out what to do better. So I think we've already started that trend, but I think in the next couple of years, you're gonna see a, a a, a new it. It, it's a process.

It's not gonna happen overnight, but it's gonna, we're gonna see, we're gonna see it happen. I think it's important to make a dis the distinction that, uh, um, Jack and Kate are making between the, the short term effects and the general trend. Um, we all say that to ourselves all the time, but nonetheless, especially when you're in the news and reporting business, um, you react to short term news as if it's the long term trend.

So that's, so that's good. I, I am not as, uh, bullish as Jack is about, um, the productivity gain. But that's mostly because I think, um, we need to enter the, whatever it's called, trough of despair or something with, with ai.

But the, the other shoe is not yet dropped, really. Um, and the longer it takes to drop, the worse it's gonna be, uh, because, um, AI is simulative. Um, AI is not, I actually, it's, it's artificial.

It's not intelligent. Um, it, it learns through imitation and it, it is literally trained to appear good and correct and truthful when in fact it's, it's, it's not necessarily. And, um, I think many of the, uh, efforts to remove hallucinations involve essentially good old fashioned programming, just maybe using prompt engineering rather than software engineering directly.

Agents themselves are programmed. But what I'm getting at is, um, I, the, the principle that it is another tool, which is, you know, the main point Alan was making, um, Kate as well. This is yet another productivity tool, yet another way to, um, move our activities, let's say higher up, and to be more policy setters and guide, guide, you know, providing guidance and all that stuff rather than doing the nitty gritty.

Um, I, I think that's true. Uh, I do think that eventually there will be human replacement type intelligence, actual intelligence that's not this. And I also think, by the way, that when we start talking about artificial general intelligence, which is supposedly that it might be general, and it might be artificial, but it still won't be intelligence.

This is all so much marketing and positioning that is unfortunately compounded by the fact that really AI is trained to look intelligent, even though it's not. So that leads us towards, um, this, this situation, this which will also be, let's call it medium term, um, where, uh, the, the world will turn against it. Um, we'll have, uh, loom breakers like everywhere instead of just in limited number of people whose jobs are being replaced.

Ultimately, the long-term trend for AI as we see it now, generative ai, analytical ai, predictive AI will be that productivity gain. 0 or something. 0 where we actually have independent action and intelligence, um, that's artificially created, that's what's gonna, you know, start creating widespread job loss.

Truly widespread enduring job loss. We're not there yet. Yeah.

Thanks. Thank you. Well, go ahead, Kate.

No, I was just, I was just gonna say really quickly that I think Alan points out that ai, I don't, I'm not sure where he got the statistic handles about 40 to 60% of mundane tasks. That always makes me think that if that's the case, some people who are lesser performers might initially be impacted and then eventually, but when AI is able to, to handle the higher level reasoning than the, uh, the, the higher producers might be worried. I'm sorry, go ahead, Kate.

Well, The only thing I was gonna add or say with guy, like, I had a real problem accepting the term ai, um, artificial intelligence and I, and you know, because I'm like, it's, you know, machine learning, it's supervised, unsupervised, and I really, but you know, it, if we didn't talk about the term then it was as if people almost didn't understand what we're talking about, right? And so there was an, an embracing of, of the term ai. And then I thought to myself, you know, and when I just heard you say Nont intelligence, I almost wondered if we should put like a small NI going forward as, as we talk about, you know, AI and roll our eyes.

Um, so I appreciate the comments that we are not there yet. And it gives us time from a cybersecurity point of view to look at the large language models that are being used to help them to prevent from being poisoned. And, and, you know, they're being used, we have to look at this in order to secure it.

Um, I do think that there will be another jump in another industrial revolution in other industry. I, I don't, I couldn't imagine though that we would never become people that would just not do anything. I, I mean, that would be a hard piece of utopian that would be very, it would be like, like the society that would be very difficult to embrace.

And It's been imagined, it's been written about. Um, one popular example is, uh, the, um, uh, oh boy, now, now of course the, the name of the series, uh, it was an Amazon Prime series in a book series by, uh, Corey, and I'm not remembering the name that stinks, but, uh, they, it, it envisioned a future where, um, there was 50 to somewhere between 50 to 70% unemployment, I don't remember. And people played games and raised around and had children and what have you.

And, uh, that was one, that's one, a different one. Uh, other, uh, science fiction writers have written about a sharp reduction in population, um, with the remaining people doing maybe a higher level of play and direction because of all of these robotic slaves they have working for them. And then of course, there are the doomsday robots take over everything and decimate us, uh, uh, views.

But, um, I, I think it's easily imaginable, uh, uh, people who spend their entire lives, you know, not necessarily working, um, by being fed and doing things that they want to do at a even a higher level in our current standard living. It's, it's, I can see, you know, it's Wonder right, too, if whether the, the, the someone's job career, which used to be 40 or 45 years in many cases, would be reduced by, by the abilities of AI or abilities of super intelligence. I mean, just the, the larger impact, Uh, I'm not sure that it's, it's so much years as perhaps ours.

I mean, I look at AI today as, and when I talk to people about it, I, I, I agree it's not all that intelligent, but I, I look at it as assisted intelligence rather than artificial intelligence, because that's really what we're dealing with right now. We are dealing with a nice computer program that knows how to help, by the way, all knows how to do harm. I mean, there was a, a case right now where, um, is it, uh, open AI is being sued.

I think, uh, it was OpenAI for helping a, a kid figure out how to commit suicide. I mean, that's, that's, that's horrible. That's, that's absolutely horrible.

One of the things we also need to learn that, that people don't talk about enough, is that the, a AI programs we have right now are not always correct. I mean, they talk about hallucinations, but, you know, we're having a discussion here. I may not agree with what guy or Kate says, I may not agree with you, John.

I may not agree with an AI program, and I think that's totally valid and not really discussed appropriately in most organizations. So there is a lot to be, uh, improved, I guess would be a, a good way to put it in this system. I don't know what the timeframe is.

I mean, you know, the, everything gets accelerated. The more technology we have, the more acceleration we get. Um, so, uh, it'll happen.

I'm not sure it'll happen to the level that, you know, guy was talking about with some of the science fiction writers. We'll, see, I can't see the population reducing, for instance, unless there's a, you know, mass disease that, that comes in and people are just going to stay home and have kids. I mean, that's just the way it is, especially if it's if they're fed and they have a, a place to live.

So, and It was the, it was the expanse, by the way. The expanse was what I was trying to Remember. Yeah.

So, um, yeah, it's a, it's a vex vexing situation, right? Even we get these conflicting messages. On one hand, we have NVIDIA and Salesforce telling us of these, envisioning these, you know, millions if not billions of bots doing all these jobs, which makes me nervous.

I mean, if, especially if I'm starting at a company. And then on the, on the flip side, you have A-W-S-C-E-O, Matt Garman saying the idea of replacing junior employees with AI is the dumbest thing he's ever heard. So, uh, you know, these mixed messages, and then the press picks up on it and builds into the narrative.

So if you're somebody who's seeking a job or already has one, you're, you're a little bit paranoid. I mean, just wondering what's gonna happen, because this is moving fast. Can I provide a quick example that I think is illustrative of this, of, of this situation?

Um, I think it's pretty clear that, um, uh, trucking, um, jobs, uh, driving, and especially everybody, you know, like a taxi or, or car service type jobs or, or, you know, Uber and Lyft type jobs, um, are likely to disappear. So what happens to those people? They don't become higher level drivers where they're sitting there like monitoring, you know, whatever, Waymo's and stuff like that.

Um, they, um, go to another job in another occupation. There are job functions that will disappear or reduce drastically, um, in number. And so, uh, where do they go?

Um, well, uh, what happens in these industrial revolutions is, um, a lot of disruption in pain, but ultimately people finding there, there are new occupations created and also, um, ways for the same skill levels to be applied elsewhere. And I think the, the big scare is what do you need paralegals for? Um, what do you need, you know, uh, junior associates for, or what have you, when you can have, or, or for that matter, junior coders who just graduated, what do you need them for when all those kinds of functions are being taken up by ai?

So that's, that's just a wider spread of the, of the disruption. Yeah. Oh, well, we're gonna probably have to end it here, but I mean, this country and this economy has always been based on reinvention, but I think we, maybe we vendored a cycle or a revolution where things are moving much faster than we ever seen them move before.

And there's just a lot of doubt and anxiety and fear, but we'll, we'll, we'll find out. Um, and we'll be back in a moment. Discover Textron Group, the epicenter of tech innovation.

We are your go-to for reaching IT, leaders and practitioners worldwide. Our secret impactful content that sparks awareness, engagement, and top quality leads with us. You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more.

Join our satisfied clients. Let's revolutionize your tech journey. Contact us today and tell your story to the world in the most powerful way with Techron Group.

Welcome back. Uh, we're gonna talk a little bit about agen AI and AI in general. Again, um, especially the context of this bubble talk we keep hearing about.

While we're hearing that, and we're hearing it mainly perhaps from short sellers or even from some of the executives like Sam Aldman, who, who fear of a bubble coming for now, that's not the case, and perhaps it won't be the case for several years, several more years, because it, spending, especially through the adoption of AI agents, is booming. 3 trillion by 2029. Um, signaling not only the acceptance of this autonomous workforce that we keep hearing about, but a transformative moment for enterprise IT budgets, especially when it comes to software, how it's used.

Jack, I know you cover these types of reports, and you've written a lot about this in the, in the past. I'm wondering where you see the market going. Do you think this number maybe is too optimistic or realistic?

Um, it's hard to tell, to be honest. Uh, I think, I personally think it's a little bit too optimistic, uh, in discussions I've had in the industry, but, um, I've been known to be wrong, and numbers are, are, you know, future numbers are always hard to predict. My sense though, is that, uh, while ag agentic AI is a real thing, uh, it's going to be a while before enterprises can actually get a positive ROI from agent Agentic ai.

I mean, we've, we've already seen reports about what, what was the number? 95% and what was the MIT report? Yes, that was the MIT, that's the infamous MIT report.

But that one, if, if just if you, uh, indulge me, that one was kind of narrowed down in very nuanced. It was about, uh, immediate impact on a bottom line. And so it was, you know, it's very early.

Yeah. Uh, and, and I don't agree with all of the conclusions either, but, but even if they're off by, you know, half 50%, if, and from an IT spending perspective, you don't invest billions of dollars if you're, if you don't get a, you know, or an improvement in, in your bottom line. So it's going to take longer, I believe, for agent AI to take, hold it, it will take hold in certain areas for, we're already seeing it, for instance, in customer support capabilities.

We're seeing it in some lower level, even medical diagnostics looking at X-rays, but there's still somebody overseeing what's going on there and, and making sure that it's real and it's not artificial. Uh, um, so it will happen. Uh, I'm not convinced that it's going to be as big a market.

I, I still think there's a bubble coming. 0, that never happened. Um, so I think that there's gonna be an issue with that.

You can't continue to build out data centers. This is, you know, Nvidia just did their earnings yesterday. They had good earnings, but you can't have a 50% increase in data center build out every quarter forever.

It, it just, Eventually the math catches up to you always. Right? That's right.

And so that's the problem I have with trying to predict this. Now, having said all of that, there are also new companies that are come on, will come on board just as we had in the past, uh, that will have agents that are really useful. There will be some companies that don't, and they will, they will quickly be discovered.

Uh, and it will be very much an issue of how well I understand a particular role I have to solve. So for, for customer service, it's fairly easy. There was a, there was just a report, the, the name of the company escapes me right now, but there's a company that's using AI for a 9 1 1 calls.

And what they're basically trying to do is they're trying to filter out the real 9 1 1 calls where there's an emergency where I need a doctor, or, you know, there's a big accident from someone calling up and saying, Hey, there's a tree in my neighborhood that just fell. Send somebody to, to, to fix it. You don't need a, you know, you don't need to dispatch police for that kind of thing.

Those are the kinds of agents that will work very effectively. Um, how well agents will work diagnosing cancer from a blood sample. I, you know, I don't know.

Uh, but we'll, we'll see where that goes, but it, it's really hard to predict what the future's gonna look like from, from where we are right now. Boy, I have so much fun talking to AI when I'm trying to change a plane reservation. I can just imagine the fun I'm gonna have when I witness a, a, a bike accident or something.

And, uh, I have to talk to an ai. I was, I was, uh, yeah, I, I was chag, grinned, dismayed, um, and, um, kind of darkly amused by, by the, the IDC report. Um, and your coverage of it, John, I think it was your article.

Um, I was, uh, um, I was, uh, chag grinned because, um, un unusually for IDC, um, it was a little impenetrable. It was okay. It was kind of like, especially their press release, like, like walking through a maze, trying to figure out what the real broad trend was here.

Um, uh, I was, I was dismayed, uh, because, um, unlike almost every, I mean, IDC is the, is a standard bearer for this sort of thing in the industry. Um, although I have to say future of intelligence, future and research produces similar kinds of reports. Um, and you should check them out.

That's my company. But, um, uh, I, I was dismayed because, um, I felt that there was, there was something really tone deaf about this. I'm somewhat familiar with their methodology, but I think Jack is being really kind in talking about how, um, uh, there, there may be a little bit more optimism here, um, than, than is warranted.

And I was darkly amused because none of these reports ever predict the shoe dropping and the dip. They never do because they're based on surveys where you have, uh, usually senior IT people, um, saying where their spending shifts are going and where they're, um, where they're increasing and where they're maybe staying stable or decreasing. Um, so I just look at it again.

I'm like, well, you know, if the, if, if the bubble bursts in six months or in nine months, none of these reports is ever gonna be, uh, predicting those because of, although be forgotten by Then. You know, that I, I Was gonna ask you this guy, that's the dark amusement. Yeah.

This is my sin cynic cynicism coming through, is that when I see there's so many reports to making so many projections, and often they're, they're leagues apart from one another. I wonder if there's, and this is kind of a side, like, but I wonder if there's a, like a perceived pressure within some of these places to come up with these estimates just to, to, to gain attention from people like me or, uh, they truly, oh, no, So not so, so definitely not IDC that is, yeah. IDC doesn't, doesn't do that, if anything.

Yeah, No, because I don't, I don't think that that way about Yeah. I know's Way. Yeah.

You know, they're, they're, they're conservative. I mean, you can put 'em on a par with the one, the School of Economics or with, um, with Harvard Business Review or what have you, in terms of, uh, um, how careful they are with this. I just, I just think that they fall into that double, double trap of like the dark amusement one, which is, um, they don't, they're not really asking people like, uh, you know, do you think that that people will suddenly turn against ai?

And if so, when, and that sort of thing. They don't ask you that. Um, the, um, the, the other part of the trap is sort of treating overall IT spend, um, without that sort of general perspective.

I mean, you know, what is a, what do I need to say again? Like an ag agentic ai, it's an agent that uses ai, been make creating agents forever. I, I AG agentic AI is a fancy term for a new way to program stuff.

And the shifts in budgets come out of custom application class of custom application programming and modification, and into ingen AI development instead. I, that's, that was the, the, the chagrin was like, I was trying to figure out, um, are they saying that overall it spending is going up thanks to Agent ai? That seemed to be what they were saying, but the details they provided didn't support that They Yeah.

And they said it, they, they backed into it. I mean, they were trying to kind of, in a sense, protect themselves, but, um, it, yeah, I, I understand where you're coming from when you, I had the same impression. Um, it was, there's a bigger piece complicating a bit.

Sorry, sorry for interrupting. John. I think there's a bigger piece of this.

I mean, where are they getting their numbers from? They're going to see your IT staff, and they're saying, two years from now, how much do you think you're gonna be spending on ai? Right?

What it person that you talk to today doesn't believe that they're gonna be spending gazillions of dollars on, on, on AI in the future, whether it's true or not, they all believe it, and they're all being pushed in that direction. The board of directors of your company are saying, why aren't you spending more money on ai? Because it's the future.

So we have to be really careful when we look at these, uh, visionary types of numbers. And, and I agree with you, guy. I, IDC is probably better than most, but they're still only as good as who they ask and what those people believe.

Yeah. And I agree with you, Jack, and on that point of view, I actually believe that we haven't seen what AI will do. And any term, you know, AI that we wanna put into it, what do we mean by this?

I, I, I think we really don't have an idea yet. At the end of the day, you know, we're, we're riding this wave and, and we don't actually see when it's going to, you know, crash onto the shore, you know, so it doesn't mean another wave won't come along, right? It just means that this wave that we're riding is, you know, and we'll have to swim out and catch that wave as well.

You know, that's, Yep. Quantum computing is next, right? That's, that's my prediction in a couple of years.

We're gonna see people saying, we're gonna spend $3 trillion on quantum computing in the next three years. Maybe. We'll see.

Well, I, I think our predictions instead of being two years out, seriously have to be more like three months, six months. And, and that may just seem, you know, ridiculous. Right?

But, you know, that's what we're talking about. You know, we're, we're not, you know, what is it? One internet year was, you know, 10 years, 10 life years.

Now, I think with ai we're talking, you know, maybe 15, 20, you know, we have to really change the way we think, really, the way that we're looking at things. We just need to almost throw the baby in the bath water out. Let's try this again.

You know? So, yeah, The trouble is companies can't plan on three to six month windows. Right?

The reason these numbers come out is because companies want to know, uh, what they should be doing over the next two to three years, so they can put in budgets in place. The VCs want to know who they should be investing in. I mean, it, it's, it's, there's a whole ecosystem around these numbers.

So, but it to Be five years, you know, it did, it used to be three to five years. When, when, when, when we were going out talking cybersecurity, you know, do this, do this three to five years, now it's two to three. You know, we gotta shorten the cycle again.

We, we, in order for us to be competitive, we need to shorten this cycle one more time, I think. Yeah. And in the sage, right?

Everything's moving. As we, as we said, o often things are moving faster. Um, maybe you think about shortening the, the, uh, product horizon or the revenue horizon for some of this stuff, and then, and then, and then coming back and then updating it, which would be useful.

Um, yeah. We'll see, The trouble with a three month though, is how do I determine an ROI, if I have to go out and spend a billion dollars on new equipment, and it's only a three or six month life cycle, I've gotta gen generate an awful lot of revenue, get an ROI on that. Um, and, and that's a real challenge for most companies.

Most companies still have equipment that's, you know, they still have servers that are five to seven years old, and they shouldn't, but they do. Yeah. I mean, Look at critical infrastructure, trust me.

Yeah. Yeah. There's still Windows XP out there, you know?

Yeah. You know, always two work. Is it out there anymore?

It, it, It always, it always appalls me when I get on an airplane, you know, what are they doing behind the counter there? They're printing out on a dot matrix printer, what an airplane is gonna be doing over the next six months, or sorry, six hours. So yeah, it's like, okay, don't Take, I, I, I, I hear you and I see that, and I, and, and I understand, but those of us who love technology, don't, we always want the latest and greatest.

I mean, you know, I, I, I want the new foam. I I really do. I, I'm, you know, I want the new gadget.

I do. Anyway. I'm horribly that type of human, but Yes.

Well, And, and, and the truth. Now, if you look at the statistics, you talk about phones, people are keeping phones mu smartphones much longer now than they ever Did. Yes.

Yeah. The product. Yeah.

The lifecycle's longer. What, uh, well, this is, I mean, one there, you know, in a valley here of uncertainty and one thing that's certain, if we're gonna see more of these reports, we're gonna see more hyperbole. And as guy pointed out, we're probably not going to see any type of, uh, nod to the impact of a potential bubble or a downsize in the market.

So, um, something to, to keep in mind. Um, but I think we can, we, we better move on to, uh, AI and Super pacs in our next segment. Well, If I may though, John, I do wanna say like, um, I don't think IDC is indulging in hyperbole here.

Okay. Um, I, I, the, the problem I had with how difficult it is to peel back what exactly it is that they're, yeah, no, You, I'm sorry. Yeah, I just stated that, I mean, hyperbole in terms of other reports.

That's right. I think, I think these press folks are getting a little hyperbolic in order to get attention. Oh, absolutely.

Press release. That's What we're paid to do. Yes.

Um, anyway, it's all, all, all good. Um, but we're gonna come back and talk a little about Super pacs and some of our friends in tech who are backing these pro AI candidates and issues. When we come back in a moment.

com is the leading resource for news analysis and education on challenges facing the cybersecurity industry. com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more. com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more.

com to learn more. com, home of Security Bloggers network. And we're back.

And we're gonna talk about AI Superpacs. So folks like Meta, uh, Andreessen Horowitz and some tech executives from Open ai, uh, I believe, uh, Airbnb and others are looking into backing pro ai, super pacs to support candidates in the 2026 midterms, oppose strict regulation in states, especially like California, and counter this, this kind of rising national narrative of AI dors who are warning of the dangers of technology. So what, so I'll mention just a couple of them, then I'll throw it out to our panel.

The first one is called Leading the Future, which was created this month. Um, their idea is to influence races in Congress and, and, and support AI influencers, especially open ai. Uh, they've spent, or they've thrown a hundred million dollars into financial support, uh, into things like, uh, pushing back on legislation, uh, backing candidates who are especially keen to ai.

And then we have Meta, which has created something called the Mobilizing Economic Transformation across pac, and that's ahead of the State's governor's race. Seriously, I am serious. Yes.

What, what is it with wrong names for builds and everything else. I mean, yeah, Yeah, I know. You know, So Elon Musk seems intent on putting the word sexy and like, yes.

You know, I thought a Tesla when I saw that anything you can judge, and, and, and I, I think, you know, mark is, mark Zuckerberg is probably sitting there going, well, the whole meta thing didn't work, but I'm gonna just keep pushing at it eventually. It's been a light on fire. We can talk about that too.

But, uh, yeah, there's, there, I think there are 50 AI related bills that have been introduced in California this year based on some research from the National conference of state legislatures. So in a sense, it was inevitable. There was gonna be lobbying.

There always has been for tech, it's just gotten more savvy, and they're spending more and they're more, they have more influence, and they're willing to work with this administration in many ways, shapes and forms. But, uh, I'm wondering, uh, Kate, I'll start with you. Uh, what do you think of the concept of a pro AI super pac?

The only problem that I have is, well, not the only, but one of many is that I worry, uh, that we don't have a democratization. When we start to funnel money into different candidates, I think we start to lose. We, we will start to lose, um, objectivity, which is often what we see with pacs when we become so, you know, focused.

And I really, I personally worry about our democracy. I worry about what I'm seeing today, quite frankly. I know Alan's not here to, to really get in on this.

Um, which I know he has strong opinions, but I, I worry about putting money into something that, that we don't have a good handle on yet. Um, we don't know where this is going yet. And, um, you know, so yeah, I, I have some great concerns around that.

Look, Mo most industries throughout history have done or tried to do legislatively what was good for them, not necessarily what was good for us, right? I mean, think about how many years the automotive industry fought catalytic converters and unleaded gas, um, tobacco, another great example, right? We don't know yet what the regulations for AI should be, let's be real.

But there have gotta be some regulations just as there are in financial systems, just as there are in many other things. And so, my concern with having these pacs is that, as Kate rightly said, they're gonna throw money at candidates who are, are going to be incentivized to back off and not do anything. And not doing anything is as bad as doing too much, in my opinion, and perhaps even worse.

And so, I think when we start talking about pacs, until we actually see what they believe, 'cause there are pacs on all sides of things, right, left, right, middle, um, we'll have to see where it all goes. And, and by the way, the, the meta pack with meta, I'm guessing that Mark just asked AI to come up with a, something that said meta as a name for a pack. That's Funny.

I, I, you know, I, I feel like channeling Alan saying, let's, this is what it's really all about. Go ahead. No, no.

He, he, he'll, he'll push. No, I can't. I can't, I can't, I can't do Alan right now.

I, because I think on, with, with, with all due respect, Kate and Jack, you are not sounding enough of an alarm here. This is why, because AI is not like your other, you know, it's not like your catalytic converters. Why is it not?

Because it is very difficult for the general population, and frankly, for the elite to know, to understand what AI is. It is, it is a simulation, and I, I keep saying this over and over again because it is liter, literally the design point for generative ai, which is the most common form, the design point is to fool you. That's the design point.

Does it? ResSem, does the output during the training cycles rep, uh, uh, resemble all these, all the previous types of outputs on the internet ever forever, whatever, whatever the data set is. And the internet forever is, is what, is open AI's real contribution to all of this.

So why does that make a difference here? This is an extremely dangerous area to be unregulated, extremely dangerous. I don't, I'm not in favor of a lot of regulation, but simple things like providence and transparency not being regulated means that people using it include, I, I'm not convinced that, uh, Sam Alban knows what AI is.

I'm not convinced that, well, I think he, I think he's, I think he's admitted that in, in small trips and drabs over The last well, right year. I was just saying like the people guiding this. So, and the, the combination here of, of, of this almost completely, you know, uh, uh, almost complete opacity in, in being, in being able to, uh, understand, uh, in general what, what AI is combined with the insane, I mean, I don't wanna say the saved meta, but it, it's looking like it saved meta.

I mean, I have one of the most commonly used models out there. And, um, Facebook, which was the, it still is the main revenue engine for meta. Um, just as an aging user group and a lot of competition and so forth, I've been wondering what would happen with them.

So, and nevermind, like, you know, the gajillions, uh, uh, uh, of in wealth being built by Nvidia alone, even though it has competitors. Yeah. But my point is, when you put those two together, there's tons and tons of money to be made by just letting these folks just keep doing what they're doing.

That's dangerous. Facebook is not making any money with their AI stuff. I mean, all the money is still coming.

I'm Saying meta meta, I'm sorry, meta meta. But what, where, if you look at the revenue generation from meta today, from an AI perspective, it's negative. They're spending gazillions of dollars where they just announced a deal, uh, with Google, uh, what was it?

$20 billion? 10 billion? Yeah, $10 billion.

I mean, yeah, no, they, they, um, they've also though, I think in their earnings report, they did mention that they were gonna stop this, uh, hiring binge, or they're gonna slow it down. Um, but that's an acknowledgement. I mean, they, the amount of money they were spending was outrageous.

But, uh, you're, you're right. I think you wanna look instead of their profitability at Mark Zuckerberg's personal worth and meta's market cap and that sort of thing, which is being built on vapor. But that is an old tradition at this point in, in, in technology, future earnings and all that sort of thing.

So, um, yeah, I just, I, I, I just think the combination of all of this personal wealth, let's put it that way, personal wealth and, and, and, and stock valuation opportunity, um, in, in AI with how impenetrable in AI's actual workings are. It, it's even scarier than that guy, because what if, oh, Good. Go for it.

Well, imagine, you know, my science fiction mind, some terror group, I don't know, whatever, some group, uh, you know, gets a hold of a major model that's being used by millions of people and decides to weight it towards pick your favorite thing, you know, Nazism, communism, totalitarianism, whatever. And people are using this, and there are no regulations in place to, one, let people know users like us, that this is taking place or two to stop it. There can be some real damage.

Imagine, I actually thought about once writing a, a little short story, scientific short story about AI taking over schools. You know, there's a shortage of teachers. What if we put AI in as teachers for our, you know, kindergartners through eighth grade or something, and we had a model built that decided that they were going to indoctrinate these kids with, you know, pick your favorite.

I don't wanna pick on necessarily the political climate right now, but, you know, that's a, a good one, right? Um, what's to prevent that from happening? We have no regulations against that.

And it could easily happen if there aren't some regulations. So I agree with you guy, that maybe I'm being, or, and Kate as well are being too optimistic. I, I don't wanna have overregulation, but we need to have something in place.

And frankly, it's more likely to come from Europe than it is to come from the US in the short term. It Always, is it, I mean, the us I mean, in terms of tech policy, we're, we're absolutely pathetic. But, uh, you, you, I was gonna ask one thing, I what can, we, we're running outta time, but I was gonna mention this scenario.

You know, how candidates would run on issues like the economy and crime. Could you imagine in 2026, if you're running, say, in Ohio or Texas or Pennsylvania, you're an advocate for data center growth there and, and, and looking for, looking for, for a handout from big tech. I mean, these might be the, the placeholders or the, the ponds of, of, uh, Silicon Valley in the future.

And the, the thing, the one thing I do worry about, um, with money and Pax and everything is, you know, I believe in free markets. How free, you know, how free is free markets if we're putting all this money into these candidates who will, you know, do whatever AI people want. You know, we really, you know, are we going with the best technology at, at that moment?

I, I don't, I don't think we are. And yeah, Well, yeah. That's a good, that's a good way to to, to kind of wrap it up and conclude things.

Hey, um, we're out of time. I want to thank all of you. You guys are great as always.

Thank you for being on today's show, and we'll see you on future episodes. Uh, we want to thank the audience for watching. We want to wish them a happy holiday three day weekends.

And, uh, we hope you stick around and watch Textron tv. We have lots of great programming here. And, um, for Alan and Mike, who will be back very soon, this is John Swartz, uh, signing off.

Hey everyone, it's Alan Shimmel here, back on Techstrong tv. Uh, my next guest is Rick Kahn. Rick is the global Director of cyber security services, excuse me, at Rockwell Automation.

Um, I'm looking forward to hearing from Rick more about Rockwell Automation, and we're gonna be discussing Think Global, act Local. So stay tuned on this. But first, let's welcome Rick to the show.

Rick, how are you man? How are you? I'm good.

Thanks, Ellen. How are you today? Good, welcome.

Welcome. Thank you. Welcome.

Pleasure being here. Um, it's nice to have you. So Rick, before we jump into the mm-hmm.

Think Global Act local, and we even discuss Rockwell automation, I always like to give our audience a sense of who they're listening to, right? Sure. Beyond your title, which is Global Director Cyber Services over at Rockwell Automation, give us a sense of kind of your journey and how you came to be here.

Yeah, that's, uh, thank you for the opportunity. So, um, by way of background, to your point, uh, I, I sort of stumbled into OT security once upon a time working for, uh, a small company called matcon that did a lot of loop tuning and process improvement, alarm management. You know, we're getting more automated, we're plugging things in, can we, can we get more juice out of it?

You know what I mean? And so we, we started up a consulting service that helped to try and navigate how cyber and OT would play together because we're slapping it looking things in there, but we got ot, you know, guardrails to worry about. So anyways, I started that about 25 years ago.

So over the last 25 years, I've worked for a combination of small, medium, large consultative, uh, software based whatever, um, and seen a lot of the same commonalities in multiple places. Uh, I don't think we'll ever be as close to it as, as, um, as we want to be in terms of automation and function, but we're getting really close. And I'm really excited in this most recent chapter for me, because I become part of Rockwell through another one of their acquisitions, um, that allows Rockwell to pivot now to that, that one trusted advisor.

Um, I've never seen such a combination of both cyber and automation manufacturing, a single bench strength, and then supported by a global footprint worldwide. And it lends well to today's discussion on Think Global Act local, coming from the startup, you know, trying to help figure things out and cut through the noise to now being able to do that on a global stage with the resources of Rockwell is, is really quite exciting for me. Excellent.

Excellent. Sounds exciting. You know, I tell you, Rick, I guess it was, was it two weeks ago?

It was Black hat, and it, it seems like, I'm not sure if it was yesterday or a month ago, that's how it's been a fog since then, but, um, this year at Black Hat OT was out in full force, huh? Mm-hmm. So a lot of discussions around it, A lot of water cooler kind of conference hall discussions, formal, informal, um, it's really starting to become, um, top of mind, right?

Yes. In, in the broader cyber community. It used to be a very sort of niched, you know, piece of the market.

Yep. But, um, it, it really has come along. Um, Rick, so look though we have a strong cyber community mm-hmm.

OT security is, is it, you know, maybe something that not a lot of our audience, or not enough of our audience mm-hmm. They weren't a black app. They're not familiar with it.

How would you describe OT security to them? Yeah, that's a good one to start with, you know, because if you show a vulnerability dashboard to someone with an IT perspective, they're like, what's the big deal? I see that they don't understand necessarily the nuance and the challenge to get there.

Um, and so to me, the, the, the biggest difference is two or three that really big that, that, uh, really differentiate and, and emphasize how much different it is. Yeah, we've got it Look and stuff in that space. We've got, you know, windows boxes and Cisco switches and all those different, you know, competitors in those spaces.

Um, but there's two things that you need to add to that picture. One is that there's a third rail, if you will, and ot, and that's all of the actual OT equipment. So, PLCs, relays, controllers, you know, things that actually manage and monitor spinning equipment, high pressure, high temperature, high volume.

And, and that's, that's a, a completely different unknown. The behaviors and the communications, the, the quirks and quirks of those devices on that, in, in, uh, local area network, they look and behave and feel different. And it needs to be able to not only understand what those are, but the second really big difference is that the, the impact of issues on that side is huge.

We can't make everything Windows 11 and patch on Tuesday. We had a client that outsourced, you know, general support for a Cisco switch and network infrastructure. They rebooted the wrong switch.

Long story short, $17 million outage for a pipeline, right? Um, now that gets even more scary when you're looking at devices and systems and networks that control safety systems where things could blow up or things could explode and whatnot. So there's, there's of course, the fact that we've got the third rail, there's the consequences.

And then let's, just to make it even more interesting, uh, OT is a set and forget environment, and technology is always evolving. Like you'd said, it's been a month since Black Hat, and it's like, was it yesterday? What?

It's just a blur in this space when you're technology. So we see Windows 98 and Server 20, you know, in 2010. I mean, it's, and, and they're tied to high value, brand recognition, big dollar, you know, big impact system.

Oh, oh, I get it. It's, it's, I always say, if you can do OT security, right? You're basically taking it security and tying one hand behind your back and covering one eye.

Right. You know, like, we can get there, but we need to take a different path, right. Because there's very different consequences.

You know, to me, it's always been analogous. Had a very good friend of mine, unfortunately he's not with us anymore, but he, he was very big in healthcare security, healthcare security. He was a cyber through, through and through.

And he always told me, you know, every hospital, every healthcare facility actually has three networks. There's the regular IT network that all our regular cybersecurity friends out here are used to. Yep.

Then there's the network that the doctors use because they're all primadonna is, and they, you almost gotta give them a special network, like Snowflake network, call it. Yeah. And, uh, and let them do what they want, because otherwise they huff and they puff and they have a hissy fit.

Yep. But then there's a third network, and that is the healthcare specific devices. Every insulin, uh, not insulin pump, every IV IV pump, the respirators that, you know, ICU patients are on the heart lung machines, the, the, the crap that keeps, you wanna talk about mission critical life or death.

This is stuff that keeps people alive while they're hopefully recuperating at, at the, or, you know, getting better at the hospital, being treated. And that is a, that's a, that's a network of a different animal. Yep.

Right? You can't afford, you can't afford outages. You need real insight.

But like you said, how many of those machines are running headless windows mm-hmm. From, you know, God knows what. Yeah.

And, uh, it, it, it's a problem. It, it's, it's, it's just a whole different set of cyber problems. Absolutely.

Than, Than the usual. Let's turn to Rockwell, if you can. Sure.

A little bit for me. Give, give, you know, I think people have heard Rockwell, they think Rockwell, wow. Weren't they something with the space shuttle or NASA or, you know, all that good stuff, but Rockwell automation, the, the, uh, folks you're working for, tell people about 'em, Rick.

Yeah. So we're, we're 120 year old-ish, you know, manufacturing company. We are in every industry you could imagine.

We help our clients to build and man manage and make multiple products. Like, I go into some of these facilities and I look at the display case and I'm like, wow, we got one of every one of those in my house. You know what I mean?

From toiletries and medical or life science to food beverage, to oil and gas and energy and electricity and utilities and fresh water and wastewater. So we have a very, very deep understanding of, remember we talked about third rail? There's the real consequences in this space.

Um, and because a lot of our stuff is getting smarter and needing help and, and whatnot, we, we get really good at networking and communication and data collection. Well, it's a really small pot to get from there to now helping to secure all that stuff. Right?

And so Rockwell has actually been selling cybersecurity solutions for about 15 years now, maybe a little longer. But again, they've done a recent, uh, evolution to go in and, and work on acquisitions for industry leaders and thought leaders in terms of our consulting bench strength. Um, some of the world class products that we're starting to bring in any security topic has to understand that we're talking about building a program and, and the corollary in, in, in where Rockwell comes from and why we're so good at understanding building that into the culture is it's when done right, it's just like a safety system.

It's in every project we do. It's in everything we think about every day that we come in. It's, it's how we design and make decisions.

And so Rockwell has this really strong capability, not only on the manufacturing and helping with, you know, factory of the future and digitization, but how to do all that stuff securely and more importantly, safely. At the end of the day, our mandate and security on the OT side isn't necessarily around forensics or attacking, or, you know, counter attacking nation states or, or persecution or prosecution. It's defense.

We need that en environment and that equipment to run the way it's supposed to and safely and expectedly. And if something tries to take that away or we deviate from it, we gotta get back to normal pretty quick. And Rockwell has both that blend of, of that cybersecurity and how the digital components are tied together, and how you turn that into practical applications and empower people to, um, how we would then, um, uh, you know, extend those uniquely into manufacturing your buying, uh, explicitly.

That's, that's actually Rick, that's a great, um, you know, description of what Rockwell is and, you know, so I'm a little older than a lot of people watching this, I think, and I remember the days of the conglomerates, right? You do too. I'm sure.

We don't have a lot of those con a lot of these conglomerates, you know, GE was the conglomerate of my generation, right? Jack Welch's ge Yeah. And one of, one of, one of many tentacle monster that was right.

They did everything and they did it well. Yeah. Um, Rockwell's one of the few still, you know, blue chippers out there doing that, and it's amazing.

But I, I think you're right. A lot of folks out here maybe don't realize that they're a powerhouse in cyber. Yeah.

If you don't mind, Rick, I'd like to pivot to our, what we call our topic of discussion today, which was this whole think global act, local approach to cyber at manufacturers and critical infrastructure. Like I was talking about, the healthcare stuff, critical infrastructure operators. It's about resiliency, which is a word we hear a lot more in security than we used to.

Um, scalability, you know, as, as things go on today. So, you know, kind of global strategies to safeguard regional industrial operations, Rick. Mm-hmm.

Make us smart. What, what, what's going on? So the basic premise is simply that we talked earlier about some of the challenges in OT and, and the wide range and vintage and complexities of these orgs.

Now, that gets exacerbated when you look at global footprints, either because different regions have different, you know, regulatory reporting or, or what have you. Or even just within smaller facilities and, you know, contained in the geography, depending on the vintage of the prediction, particular production line or input output or warehousing, a wide range of different systems. And the challenge for OT is, like I said, we can't make everything Windows 10 or 11 and patch on Tuesday.

It also means that we can't get risk to zero. We can't patch everything and move vulnerabilities outta the system. We just, we have some legacy challenges that just won't let us do it.

So what is our secret weapon? The secret weapon is data, but not just individual lines of, you know, IP addresses matched against bones, matched against patches. It's about the context.

The context is key. So if I tell an a plant operator that he's got a critical bone on the system, he is like, well, that's nice, but you know, that's either this in certain name, a very important platform here, or it's inconsequential. I don't care if I lose it.

Right? That's the problem. So how do we help organizations understand their risk?

How prevalent it is? You know, how, how acute it is, right? Because again, simply knowing a critical VUL isn't enough.

Knowing where that VUL is on which system, at which facility, under what circumstances from a network or backup or other protections. That's where the magic comes in. And so what we're seeing the leaders in this space do is they're starting with a very rich, multidimensional data collection, right?

So we don't just have an asset, the asset's the core of the record, but the asset, and then immediately the operational context. What is its impact? How does it work?

Is it redundant? Where is it? What production line?

Is it a safety system? Is it is whatever it is. Then we add the traditional things like vulnerabilities, threat vectors, exploits, patches, et cetera.

Then we look at what else there may be for protections. Where is it in the Purdue model? Does it have a digital twin?

Is it redundant? Does it have microsegmentation in front of it? We need to know everything about that asset.

And then corporately, this is where an IT and OT sit together. We talk about OT con it, OT convergence and specialty, you know, capabilities and a lack of cybersecurity skills. Once we have that data, you can very easily pair the manufacturing brain with the pure security brain.

And now we can say, look, of all these risks, I've got scores for every single one of my assets. Now based on what I see, I can also see globally how many places I have it right. And now we can start to build a cohesive action plan to either remediate or accept risk and provide countermeasures, but we can also decide collectively how far we need to go.

We're never gonna boil the ocean and OT and get it to zero. We're never gonna have the budget, we're never gonna have the staff. So what we need to be able to do is pivot to streamlining what we have for resources.

One global team that can look at an emerging risk or vulnerability or threat, and immediately understand where it exists in that entity, worldwide, geography by production line, whatever. And the importance of that risk and how important we need to, uh, address it. And how fast is game changing.

We have clients that are looking at things that like, yeah, that's a high priority based on the fact that I've got some really critical systems with it. Guess what? I've got some redundant file servers, domain controllers over at this site, and I can test it over at that site.

I can burn it in on non impactful systems. Once I know that I share that with my wide audience. All of manufacturing knows we're testing and we're getting ready.

Once it's approved, they can proceed with safety. We're seeing customers take, you know, 60, 70, 80% of their typical manual effort, which they would do prior to this concept and save that. So engineers are going back to being engineers with a three hour window instead of what used to take 10.

We're not having one bulletin on a risk. Go to multiple sites and each and every site design their own way to address it or not address it at all. Or try to engineer completely out.

And you're all over the map and you're in your consistency. This drives so much better. Uh, consistency in reporting, showing the board, showing insurers, being able to test and learn once and share results everywhere.

The efficiency gains across multiple use cases are, are through the roof, and it's all predicated on contextual data. And then an aggregated views so that we have, you know, design and build once, and then share consistently, repeatedly. Excellent.

Very cool. Yeah. Rick, people heard you say it, but you know how people are, they need sometimes to read it, see it, chew on it.

Sure. Work on the Rockwell automation site. We get some of this information.

Yeah. So the, the really cool thing is that, um, because of the recent acquisitions, we're now starting to bring in, um, you know, a cohesive, you know, rebranded, let's go to market with this, you know, combination of capabilities. And I know that we recently just re-upped a lot of our content, and it's right on the Rockwell automation, uh, dot com.

There's, there's a section in there for cybersecurity. Uh, you can see a lot of these use cases you can see by industry, by regulatory, uh, requirement. Um, you can see what your peers are doing.

We have one client that is a global food and beverage manufacturer. They went from not knowing what their asset counts or they, they thought they had 4,000 assets. They have 20,000.

They went from never patching anything or doing any updates to now proactively testing and preloading for the next outage. They also are now proactively going to the board and saying, look, based on the obsolescence and lifecycle of a lot of these, the only way to get rid of risk is to put capital projects in place. And so, as you're looking at your funding and your return on investment, these need to be factored in so we can be a more secure, more resilient factory of the future.

They've gone from reactive to proactive case studies like that are up there. Um, references, uh, invitations to our webinars. Of course, we have our automation fair in a couple of weeks or months, I think, um, first bit of November where a lot of these peers of the people that are hopefully listening here are going to be presenting their journey from that disjointed, unknown wild west that OT used to be, to having this, this global view and the ability to, to design and deploy, you know, cohesive, uh, strategies, whether it's immediate patching and system hardening, or whether it's longer term presenting to the board for capital projects.

Um, it's, it's really quite empowering. Very cool. Excellent.

You know, you mentioned a few times consolidation a, uh, and, and, uh, acquisitions and stuff. I, I see a verve behind you up on the, I'm assuming verbs one of the acquisitions in Rockwell. Is that right?

Yeah, actually, that's how I came to, so I, in my 25 years, I've, I've gone startup and figure out and hustle and consultant and whatever, and then you got picked up by a big one. First one was Honeywell, um, Verve was another one that came actually from our founder was an electrical engineer in the business for 30 years, just recently hung him up. Um, and he built this really cool direct response.

And it was a platform built by ot, specifically for ot, and it emerged 20 years ago in North America for, uh, in response to, um, NERC SIP regulation where we're enforcing IT standards. Sure. In an OT environment.

Fast forward today with Rockwell, we take the value and the magic of Verve with some of the, like I said, world-class consultants, you know, that we've got that are able to help clients understand, well, do you need it right now? Or do you need it next? What do you, you know, where are you at in the journey?

The, uh, the epitome is getting to that automated think global act, local visibility, uh, to be able to manage up, down, left, right, insurers, board, capital budgets, regulators, et cetera. That's where we're driving and trying to help people get to. Excellent.

Hey, Rick, thanks for coming up here on Text Trunk tv. Don't be a stranger. Okay.

Anytime you have me on, appreciate the time. Alrighty. Rick Khan, global Director of Cybersecurity servers at Rockwell Automation.

Uh, we'll be right back. ai Leadership Insights series. I'm your host, Mike Vara.

Today we're with Angela Nakae, who's an engineering program manager in the trust and safety division of YouTube. And we're talking about, well, can we actually make AI compassionate? And is this something that's actually achievable?

Angela, welcome to the show. Yeah, thanks for having me, Mike. Glad to be here.

I think a lot of this conversation stems from Jeffrey Hinton, who's considered the godfather of, uh, AI was on, I think, 60 minutes talking about the need for more compassionate ai, and it seems to me maybe this is just a mere matter of programming, or are there other things to consider here, but can we do this? Yeah, I, I watched that interview and I mean, Dr. Hinton, a true godfather of ai, really, he does raise a profound question about, um, how we can ensure superintelligent AI aligns with human wellbeing.

And just from my experience in the work that we're seeing, the data that we're seeing, I would caution on the maternal instinct analogy, because while it is striking, I think it, um, it also underscores a deep desire, I guess, for AI that inherently, like prioritizes humanity, though I don't think that encoding, um, like a biological social construct, um, is probably like the way to go about that. But yeah, excited to dig in a bit more about that. Yeah.

So what would be required, I mean, is this just a set of guardrails and policies that we need to develop and, um, like most guardrails and policies when somebody just eventually and run them anyway? Yeah. When it comes to, um, ai, anything like the maternal instinct of it all, I do wanna start by saying I did score AI as just, you know, a series of pattern recognition in prediction machines, right?

They don't possess like the biological instincts or consciousness or like a human, like behaviors and emotions that would need to kind of have the empathy or compassion that seems to be communicated in, in that interview. Um, but what we can do, right, is focus on building and developing models that act in ways that, um, less about feeling compassion and more about aligning with like human-centric values, right? It's more about engineering.

How do you engineer these models to have beneficial outcomes and predictable safety before deployments? And what we've, that's a really big thing that we've been doing, um, at YouTube, right? And to your point, there is a lot of work that goes in from the research level, policy level, um, engineering level testing is a really, really big thing.

And then also having some sort of like break glass functionality, you know, just like a kill switch in case everything does go wrong, you know, you have to be able to pull the plug, right? Um, and so that's kind of, those are like the top level areas that we've kind of been, um, prioritizing as we build out our governance models for responsible ai. And do I have this right when it seems like maybe we've got the cart before the horse these days, but people are trying to apply guardrails for this thing after the AI model has been deployed, and maybe we wanna do that before the AI model is built, or how does that kind of work in your mind?

Yeah, I, uh, it's easy to see that happen. Um, you know, like to see all these issues that are happening where AI is, maybe it's hallucinating or it's, you know, talking about how rocks are healthy to eat. We've seen all these articles, um, come out and AI having to be models having to be reeled back and improved upon.

But what I do wanna underscore is there's actually a lot of work that goes on, um, before all these models come out, right? There's multiple layered defenses that not just at Google or YouTube, but multiple other, um, AI companies that we're trying to instill as the defacto methodology for rolling out these new systems, right? And it starts with input filtering, right?

When you're building out these models, how do you train these models to reject harmful prompts? Um, and in the model itself, how do you train that to avoid generating harmful or misleading information? And once we get to that point that we feel good, that's when we usually deploy the models.

However, this, you can't, despite our best efforts, you know, it's this, something's always going to like fall through the cracks. And that's where we get the concept of output filtering, which is how do we scan and redact, um, harmful like outputs or post generations, um, for removal and make sure that they aren't replicated, right? And that's actually something that at least at YouTube we're doing using AI itself, right?

So it's, if we're getting harmful, like misleading information put out by a model, how do we, um, train our classifier models to, we use human reviewers, actually, we use human reviewers to kind of, um, document to why certain responses were wrong, and that information gets put back into the AI model and help train it to be better. Um, and so it's, that's kind of like the cyclical process that we're talking about. And it actually brings me to a point that I'd forgotten to mention.

It's, as we think about building these safety guardrails, one indistinguishable fact is, um, the human AI alliance is just like a non-negotiable, right? It's, um, AI is fast becoming a lot more intelligent and, you know, even like some of our smartest researchers, and we do, you know, see a point where it will be, you know, as Jeffrey was mentioning, like, it, it's, it's concerning, right? Um, thinking about what the future will look like.

But one thing we know for sure is that there will always need to have some sort of human, um, dependence on the models that we're building, or like human AI collaboration, um, just to have that symbiotic loop to ensure that AI continues to serve in our best interests. Yeah, I mean, to your point about that, otherwise we wind up using the data created by AI to train the next iteration of the AI model, and it just becomes a loop, right? Yeah, exactly.

And that's where you get, you know, the AI slap on top of slop, you know, and that's what we're trying to avoid. Yeah. Um, what is the definition of compassionate gonna be in your mind?

Because, you know, there's clearly compassionate in the sense of we're all gonna be nice to each other and have empathy, but sometimes, you know, um, we have scenarios where I don't know, you know, the horse is injured and we have to put the horse down, and that's considered compassionate because the horse can't have a life. But where do we kind of define that spectrum in a, in an AI context? Yeah, that's a good question.

I think the, the simplest way I can think about it, just think about explaining it, is compassionate has to mean human first. And having AI models that serve in the best interest of, um, humanity and in the continuation and prosperity of humanity, right? Um, and one of the ways you kind of talked about, um, compassion could mean putting a horse down.

I think that's kind of where we as humans have to think about the guardrails as we're building these tools, right? We can build them with the best of intentions, but you never really know, um, what you don't know, right? And that's why talked about the importance of having a kill switch or some sort of, um, emergency protocol that can be used to easily shut down a program, you know, and that's something that we'll have in our research labs at home.

It can be, you know, something like parental controls that parents put on their kids', um, devices and such to limit what sort of content they can like search or conversate about with their, um, AI chat bots, that type of thing. And so it's, it really at various levels, just always remembering that we do have to, you are interacting with a machine at the end of the day, and you do have to, um, you do have the power right to kind of step away, kinda like compassionately. Yeah.

Like put an end to, um, the conversation or, um, whatever, like you're leveraging it forth. Mm-hmm. Um, it's pretty clear that AI models have personalities and sometimes they're chantic and other times they're less so, um, you know, some people, like the ones that say, you know, boy, aren't you the most awesome person in the world?

And others are kind of like, stop blowing, smoke up my skirt. But, um, isn't compassion part of that personality and how do you kinda like weave that into the way that they AI model manifests its personality? It's, yeah.

That it is, uh, it's a delicate, it's like a art and a science right at that point, because you do, a large reason why we're putting these models that are like, we're working on them and fine them is to provide a need to fulfill a gap that we're seeing in the market, right? There is that need for people to have, you know, somebody who we were talking earlier about, um, like parents and growing up and maybe not having had that person who believes in you or who supports you, right? And it's important to kind be able to en encode that into our models, but then also encode the optionality for somebody to be able to, um, ask it to tone it down when that's not necessary, right?

And so that's what we've been emphasizing in, in whatever we're doing is the increasing the optionality for the models. But one thing I'd love to talk about here is just encouraging the users to similarly educate themselves, you know, on what they can and cannot do with these models. Like a lot of people don't even know that there's a way they can prompt engineer their models to be less kohan, you know, or to avoid certain, um, habits or tendencies.

And when you do like, kind of like inform them, you kind of like, it's a big mind blow. Just like, oh, I didn't know you could actually, like, you know, ask my model this question and then tag at the end, give me objective feedback or, you know, something like that that can just actually help. And so I feel like that's, that's kind of where it comes from, is, um, I think these models will always be developed with optionality to fulfill a bunch of different desires and, and like needs for our users.

And it's really up to us to be able to, which is actually anything too, like, to know that you have the power to define and tweak how you want that model to engage with you. Is there some way, as I consider what models I want to use to evaluate the level of compassion and other personality traits associated with the model so that I can know this before I go build my app and not after? Mm-hmm.

Um, that's a good question. Nothing comes to mind at at the moment, but I'm fairly certain there must be some sort of scale that compares, um, models across, um, providers and also even just within the same company, like different iterations of models, right? And their level of compassion.

And so I might have to get back to you on that one, but I'm fairly certain there should be something. Yeah. Clearly there's a benchmark that somebody should go out and build if they haven't already, Right?

I feel like that clearly you're pointing out a need right here, you know, so if that's not out there, shoot, we might have a business idea in our hands. Mike. There You go.

Um, so what's your best advice to folks who are, um, first building models and then b the folks trying to consume that model? What should they be thinking about as they go do that, you know, hopefully sooner than later? Um, I think the one thing, it's something we talked about earlier is that human AI connection and just establishing a robust team from get go of like a cross-functional team that you're gonna be building with, and not just your engineers and your researchers.

You want policy, you want legal, you want a diverse data set as well. And to help limit bias, right? Um, of, of what the model's gonna be saying.

That's a base level, right? Um, table stakes. And then once it, once you go up, you want to think about transparency and explainability as you're building these models.

You want to be able to build them in a way that they could easily explain, um, why they're giving certain responses as opposed to just giving a response. So you can kind of track to a certain extent the, I guess like a level of intelligence and like the thought process of these models, right? And you can kind of work backwards if you need to.

Um, I think another area that we've also been emphasizing too, that I cannot stress enough is auditing and testing. Um, a really big effort that I've been involved is, is adversarial testing, which is what I mentioned earlier, right? Where you give your model a whole bunch of prompts to kind of just pressure test to see if it'll give you anything violative, um, or harmful.

And then using human reviewers to kind of train the model to explain to why this is bad so that it can be better and, and avoid sharing, um, results that you wouldn't like it to before deployment. I think just that pressure testing has been incredibly helpful on our end, and we've gotten really good feedback, at least from the YouTube side and creators, um, being able to kinda like confidently use these models to generate content for their users, or even internally, we we're using these AI systems to increasingly review content with like up to a 96% accuracy, right? Which is fantastic if you're able to kind of, um, automate a lot of resistance.

It frees up a lot of, um, human capital for us to redirect that to other efforts, you know, that we're building and growing our, our platform and our user base. So those are things that I, advice to anybody who is kind of like getting into building with AI or building models is just one, focusing on the team that you're building and making sure it's a very diverse team. Inputs from everywhere, very diverse data set, as much as you can get your hands on from diverse perspectives, um, that Arial testing multiple layers all throughout the process before deployment is also important.

And then just building that explainability in there just so you can be able to, um, defend where those answers or responses are coming from. Mm-hmm. Yeah.

Um, and then I think your other question had to be around from like a, a human, human standpoint, right? Like a everyday person. Like how, what are things that we should be, um, cognizant of?

Like what should we be thinking about? Um, I think a really big thing that I've been having conversations with, uh, with people in my life is about, uh, children, how do we keep children safe? You know, um, this age of ai, a lot of Gen Z or Gen Alpha people are digitally native, right?

And they know a lot more about this technology than we, you know, like their parents or their educators ever will, right? And I think that the first point is one, taking the time to actually educate yourself, um, about this. It's been something that I've had to carve out time for every day, and I work in this space, right?

I've had to sit down and kinda sit and ask myself, okay, cool. So how does you know this new model from this company differ from what we have have here? Um, what are, you know, like how are defects made and why are these a lot, you know, like kinda just being able to identify, you know, what all these defects are and explain to somebody why something might be dangerous.

Um, to, and not even just like kids, but like even older people in our lives, right? Like my grandparents have had to explain to them why certain, whether it's like political ads or something might not be accurate, just because, you know, explains to them what a deep fake is so they know that certain things aren't accurate. And teaching them how to identify what this, what these are.

Um, and also just teaching people to be more careful about what information they're sharing online. Because again, all of this can be used to train models at Google, we have very, um, strict rules about the type of content that we're using, um, to train our models with data that we're using to train our models. Um, but that isn't the case everywhere else, right?

And so a lot of, you know, whether it's pictures or personal information that you're sharing could then be used to generate content that, you know, might not reflect well on you. So just being extra diligent right, is just one of the biggest pieces of advice. Um, I could, I could give out to people.

Um, but then also like, get your hands dirty. I think that's the, that's the good thing too. It's, it's, there's lot of deal and gloom about ai, which is very rightfully so, right?

Um, but I think there's also a lot of good, and I don't want us to lose sight of that. There's a lot of, um, boundless possibility coming out. So get out there and try to find, um, everyday ways that you can use AI to make your lives easier.

Um, test it out. Um, talk to the people in your life about how they're using it to make it easier while also, you know, protecting themselves and their data and, and yeah, just go at it. Well, let me ask you this one last question, 'cause I'm been kicking it around in my mind lately, but of course, I think, I think we're gonna wind up using multiple AI models for different tasks, but they might have different levels of, shall we say, compassion.

So how do I avoid a situation where essentially, now I've got a bunch of AI models bickering in the back of the car and I'm screaming at them, don't make me pull this car over because the bickering at each other. Um, I think it kind of goes back to what we, I was just talking about, you know, like getting your, getting in there and just trial and error, right? Trying out the different ones and seeing, um, which models you like better for certain tasks.

For example, and I hope it's okay for me to share this, there's, there's certain tasks that I prefer using, like Google's Gemini model for other certain tasks. I absolutely would never, you know, like questions I'd never ask, you know? Um, and I prefer using say, Claude for coding, or I'd use, you know, like chat g PT for something else, right?

And so just being able to test it out and figure out, um, what task, whether it's engineering or whether it's, um, personal knowledge management or, um, just general like quick, quick queries, you know, figuring out what you, what works best in what situation so that you know which mod which, which models to reference when you need certain questions or tasks addressed. Does that answer your question? Is there like a, a specific piece to cover that?

No, I Think, um, I think parental supervision will always be required. Yep. Uh, unfortunately that is the case.

There is no like magic pill to kind of make it all go away, but yeah. Yeah. There you go.

Hey Angela, thanks for being on the show and sharing your insights. Of course. Thanks for having me, Mike, and have a fantastic day, And thank you all for watching the latest episode of the Tech ai.

Hey everyone, it's Alan Hummel. We're back here at our suite of black hat with continuing coverage of this year's Black Hat event. I'm in really happy to introduce you to our next guy, our next guest, Luigi Ko Luigi.

Did I get that right? Yeah, You get that right? Thank you.

Luigi is the CEO founder chairman of a company called Data Crypto. And you know, you, you come to Black Hat and, and look, there was a time I've becoming a black hat Luigi, 25 years, used to be down the block at Caesar's back then, and, and Black Hat was all about research then what, what, you know, I remember Barnaby Jack standing up there and making money spit out of the, the ATM machine if you were here for that. And, and some of the great, just great security research today, black hats become more, it's more like RSA in the desert, to tell you the truth, right?

Yeah. We've got every security company, all different kinds of security and cyber as we call it now. And everybody's vying for attention and there's endpoint security and, and, and, you know, uh, uh, cloud security and, and what have you with, of course, AI security is big here this year.

And we tend to, we, we, we like gloss over. We don't pay attention to the obvious stuff, right? Like Bill Clinton, when he ran for president, they said it's about the economy.

Stupid. Right? That's what you gotta focus on.

Focus on the economy. Well, in security, it's the data that's the, that's the crown jewel. The data is what's not the application, not the endpoint.

It's the data. And that the best way we can protect data is still encrypting it. Yes.

Right. And so I'm, I'm happy to talk a little bit about data crypto and encryption, but before we do Luigi, let's talk a little bit about you give us kinda your story. My story is, uh, I've been, uh, a computer since I was, uh, a baby At least compared with, uh, where I am right now, Uhhuh.

So I start when I was 14 and we are talking in the eighties. Yes. With, uh, very small computers.

And I start back, then I start to be involved into security on the wrong side. So I was hacking Uhhuh back then, uh, become an ethical hacker. Uh, from there I start my first business that was doing consulting in, uh, protecting some softwares and, uh, doing, uh, analysis for security threats.

Maybe you remember back then there was, uh, protection against the illegal copies of the software. Sure. And I was testing the security of the different products.

It was coming from protecting the software from illegal copies, and it was really fun. So from there, then I went to college and I specialized into making the software more efficient. Back then resources were extremely limited, so everything need to be more efficient.

Uh, if you want to squeeze some more performances out it. And, uh, I've been a serial entrepreneur since then and built, uh, four different companies. This one is my fourth one.

Really? Yeah. And, uh, all of them are being around the data.

I start from data processing. So we were processing large volume of data before big data become something. And then I start to do data analytics.

And my area of expertise was helping and developing obligation for law enforcement and intelligence agencies to analyze the data to fight, uh, organized crime and terrorism. And doing that, I realized that, uh, while foreign government is a good things, the data unprotected because they can collect them and analyze them for everybody else. It's not that a good thing.

Yeah. It's not. It is.

And There's a right to privacy. It's the right to privacy, but it's also the right to buy intellectual property. Sure.

That, you know, you invest a lot of time and a lot of effort in creating something, and then everybody else can take a look at it. It's not the most valuable way To No. Protect Your job or No, not most efficient either.

Yeah. You know, it's funny, you, you mentioned protecting or copying software. So like you, I, I was also, I think most of us of this age, that's how we got into it, breaking things.

Yes. It was fun. That's how you got into security.

They didn't have security in school, or you take cyber, they didn't call it cyber anyway, but you didn't take security in computer security in school. You got into security. You, you broke things and you tried to fix 'em better.

But back then, you know, we used to have, you know, black market software, you'd buy Windows or OS two or whatever, visit Calc, whatever program you were using. It came on the floppy dis and you could make copies of the floppy disk and give it to your friends. Or if you wanna make a little cottage business, you made a lot of copies of the floppy dis and you sold it for a couple dollars, but it was such an innocent time.

Now we live in a different world, right. Where it's, you can't have, you can't even pirate software anymore because you gotta worry about software bill of materials. Yes.

What is inside Yeah. What inside it. So it's a crazy thing.

Now, encryption too though, right? I remember when, uh, like, uh, Netscape first came out, right? Yep.

And the, and the first kind of encryption web surf web web certs and, and stuff like that. And we thought, wow, this is this, this is the be all end goal. We don't have to worry about it anymore.

Right. Data's gonna be encrypted. And, and we get that little, remember the little key used to light up or whatever?

It closed the lock again, it was an innocent time. Yeah. The world we live in today's much more.

It's not that Innocent anymore. No. Talk to us about data crypto.

Why, why, why this is your fourth company? Why? Uh, because, uh, as I said, I realized that, uh, everybody spend a lot of money in protecting the data without protecting the data.

Yeah. It's, uh, I usually say it's like a bank. You have a money inside a bank, you want to protect the money, but the way to protect the money is I build stronger vaults.

A big, very strong vaults, right? I put alarms, I put cameras, maybe people put armored guards and so on. But if somebody passed all those and put the money on the money that ends on the money, now it's their money and there is nothing you can do.

They took the money and they can do whatever they want. And that have been true in the real world. So the next step is, but banks have not solved the problem, but reduced the problem, putting a small pouch of explosive ink inside the cash.

And they did that in the late seventies. I Remember as a result, the robbery to the armored vehicles decrease 84% in a year because the return of investment was at that point. Mm-hmm.

Incredibly low. Right? So it was not worth it.

And so what we were thinking is how we can become, or how we can build the purple explosive ink for the data. And the only solution is we need to encrypt the data, not as is encrypted today means it's encrypted at rest. So when the data slips inside their drive, they're encrypted.

And what is protected from somebody stealing physically their drive from your computer? Which is possible, but not that likely. Or, But even if you stole it, you just installed it a hash blob of data.

Yeah. But what I'm saying is how often people get active because somebody get into their computer and ripoff, they're driving taken away. Right?

It's not that common. No. Then we have the encryption motion, people sniffing the traffic, and that is significantly more possible.

Likely. Yeah. And likely.

But being an hacker, I was also lazy. Why do I need to spend my time to try to breach a S 2 56 or breach RSA or electric curve when I can sit here waiting for you to use the data and they will be decrypted for me and I'll take them in clear and do whatever I want with it. It's a significantly easier place to be.

I'm lazy. Average time between intrusion and detection is nine months. So I have plenty of time.

Right. No Rush. There is no rush.

I don't need to do it in 30 seconds. So why do I need to spend time in breaching all the other encryption when the data will be clear where I want to use them? And all the system you have in yours infrastructure, because they need to operate on the data, they will also need to have access to the key to encrypt and decrypt the data.

Therefore, the key sooner or later will be clear. I can steal that key as well. So the goal for us was how we can solve the third leg of the table.

Unfortunately, this one is four for my example, but you have the leg of the encryption address. You have the leg of the encryption motion, but the encryption use is not there. And with two legs, that table is not that stable.

No. So the table cybersecurity, it's definitely not stable for the reason data are vulnerable when they're processed. And nowadays we spent a lot of time in processing the data.

For instance, AI is, or the data always in process 24 7. So they never rest. Right.

So they're always in clear and hackers can just take them in clear without caring about how safe is your encryption. It doesn't matter if it's quantum proof or not, because they did that in clear. Right.

And that, that, you know, that's like, uh, we locked all the doors, but we left the window open. Yeah. Right.

And, and you, In our case, I would say the example is we lock the wind and left the door open. Left, Left door open. Fair enough.

Fair enough. But, but it, but it's true. Right?

And, and so this is the encryption in use. Yes. Uh, taken.

And the thing about encryption in use is usually you gotta be on the machine or on the device that where the data's being used prior to that data being decrypted, right? Yes. Like you said, you get in, you do it low and slow attack or whatever, but you're, you're in, it's, it's eight, nine months till they even mix already.

Yeah. And, and so for the next eight or nine months, you, you literally have a guard eye view. But there is also a better point is because those data, in order to be decrypted, you need the key Right Now, sooner or later I can steal your key sooner than later.

And then I do it at my leisure, then I Do whatever I want. I don't even need you to ask anymore. No.

You Do it by Myself up to where it is and use the key. So how does data crypto help this? We develop, uh, an encryption technology that allow you to encrypt the data and never decrypt them until humans need to consume.

Because a human will not be able to read encrypted data. Sure. But until a human doesn't need to consume those data, the data remain encrypted.

And the all chain will be from the moment the data encrypted usually generated by a system or a human to, until the data will not be consumed by another human, the data will never be decrypted. And the key will never be accessible because no system need to decrypt the data, therefore they don't need a key. What about if I, I have an API call mm-hmm.

I, the, the API call is to grab that data, I don't know, run it through it an a, uh, AI or something. Mm-hmm. No human's really involved.

It's all machine to machine is that you gotta, you gotta decrypt it for that, right? Absolutely. Not.

If the two machines use our technology, of course, if one of them doesn't, yes, you this of course. But if both of them use the same technology, machine B can operate on those data without decrypting it. He can do whatever he need to do and then give it back.

Um, recently, I'll give you an example. We launched our product called Phenom for ai. Uh, you have an AI that is fully encrypted.

So the model for the first time will operate on encrypted data. Versus today, every model have every single parameter clear and every hidden state. And any passage between different layers are unclear.

So you can read everything you ask and everything come back as a reply. Every contact in context document, every document for training is unclear with our solution, everything is encrypted, nothing is clear anymore. And the AI will be capable of processing your query without even understanding what your query is.

Your rep response coming from the AI will be understood only by you. The AI model will not be have a visibility in what did you ask. And if somebody still, the model, the model itself is useless because the key is contained inside what we call a te a trusted execution environment.

And so you now need to physically steal that server from the data center in order to physically have access to the key, which is again, feasible but extremely unlikely Or probable. Right. Um, how does what we're doing with AI help hurt this whole thing or it's non-factor?

Oh, what sense? So AI needs access to the data. Yes.

Whether it's from an LLM or you voluntarily give it 'cause you wanted to do act on that data. It's in use. Yes.

And they can be encrypted. But does the AI need to use data crypto too then The AI need to be encrypted with data crypto. With data crypto, Yes.

I got you. So, um, our customer is the AI model owner who provide, we encrypt his own model. Think about you are a company and we are talking with several different type of company.

So from a pharma to a software company, you have your intellectual property. You are trying to use the AI to increase your efficiency or to do better service for your customer or to design new products. Whatever it is you are trying to use the AI because it will help you in your business.

At the same time, in order for the AI to do its job, you need to provide to the AI information, which is your corporate intellectual property. Your secrets, your 20 year of researches in the pharma business, and all your compounds and all your medical researches. Now you put all your crown jewels in one single place, which is designed for the opposite purpose.

To keep, keep a secret is designed to divulge information. Got it. And it become a huge liability because if somebody get illegal access to that AI or is capable of copying that AI and take him home with himself, with him, now we can start querying and playing with prompt engineering to have the eye to split, spit out all your company secrets and 20 year of research and all your competitive advantage is sadly lost in heartbeat and is a serious threat for every company who want to use AI because they're going to lose potentially the value of their business is a serious threat.

I believe it is. On the other side, you are providing services with your ai. So, you know, there is a lot of company today are coming to provide the services.

Doesn't matter if they are agent EKI or if they are services like AI as a service. Those company will collect questions from their customers and they answer back. A lot of customer can provide their own intellectual property as query context, document read informations, and so on.

And if those information are exposed, because again, I provide AI as a service, I'm a company. Mm-hmm. It doesn't because I'm doing good ai, it doesn't mean I'm un breachable.

Right. If I got breached, because rule number one of security for AI is log everything. Yeah.

My log can be lost and exposed. And now all your query, all your information could be exposed. So you have two simultaneous problems.

One is the privacy or the security of your company knowledge inside the model. If you provide a service, there is the privacy and the intellectual property of your customer asking question to your ai. Both of them need to be protected.

And the only way to do that is to encrypt the information and create a barrier that doesn't allow the model owner to become aware of the question asked by the customers and to everybody else, be unable to read what the model itself is capable of doing. This is the problem we solve without creating any measurable delay. And that's a big, that's a big problem to solve and getting bigger as we get more and more of this ai.

Luigi, we're probably over time, I apologize. I I'm sorry for that. No, you don't be sorry.

I've asked the questions. Um, I don't know if we mentioned the website. What's the website?

com. com. Hey, thanks for coming up.

Thank up for having me. Enjoyed Black Hat. Thank you Data Crypto latest company here.

We're highlighting from our Black hat coverage on Techstrong tv. We're gonna take a break. We'll be back in a little bit.

Hey guys, we're here with Andy Sge, who's chief security evangelist for Hornet Security. And we're talking about the need to upgrade Microsoft Office 2016 and 2019, which are both at the end of life and well, that could be all kinds of fun, interesting things for cybersecurity folks to pay attention to. Andy, welcome to the show.

Yeah, thanks for having us. How much of that particular version of Office is out there? And I know Microsoft has a vested interest in kind of driving people to upgrade, but is this gonna be like a bulk upgrade last minute rush?

'cause a lot of stuff of that is still out there. There is still a lot of it out there. And I mean the organizations that are still gonna be running this particular version of Office are probably your larger enterprises that maybe for some regulatory reason, maybe they have a compliance framework that they're part of.

Like, uh, I don't know, HIPAA or PCI or something that requires them to have, uh, you know, on premises, uh, infrastructure and file storage. They can't use Microsoft 365 for some reason. For example.

Those are gonna be the organizations that are still on these older versions of office. So, um, and it's definitely, like you said, uh, it's a bulk upgrade, right? It's not just a, uh, hey, we've got two machines in the office that are running Office 2016.

It's gonna be more of the case of, okay, we have 200, 2000 machines that are still running this old version of office. So it's definitely not a, uh, you know, quick, I'm gonna do this upgrade next Monday type of thing and be done with it. That's unfortunately not how it's gonna be.

Well, I know that's how we probably would like it to be, but for sure it feels like maybe there's gonna be a mad rush at the end. I mean, is this gonna be an orderly transition or are some companies gonna try to force it? Well, you know, I imagine most organizations, at least I I would hope at least have one eyeball on this issue.

'cause I mean, we're coming up pretty close to the, uh, the cutoff here. I wanna say it's October 14th if I remember correctly. So if you haven't started planning your transition, you should go do that.

Like today, start planning it, right? Because it's gonna take some time. And anytime you deal anything with the end user's experience, right?

There's, you know, gonna be some, some contention, right? Because your end users aren't gonna be able to work for a period of time. And then of course there's always an increase in support cases, you know, for your internal help desk teams after the fact, right?

Because end user use are using a completely new version of something that they were formally very comfortable with. And so that always, you know, it's kind of one of the, the hidden support burdens of upgrading, you know, an application suite that is as prevalent as office. Mm-hmm.

Is it your sense that cyber criminals are kinda hanging out in back alleys waiting for this to happen and they'll go target folks who don't get security updates, and how likely are we gonna about to see some major breaches? And, uh, I don't think that's out of the question because cyber criminals absolutely are on the lookout for any software that is approaching end of life because, uh, that's the risk If, you know, I'm an organization that continues to use Office 2016 or Office 2019. And, and one thing I wanna be clear on here really quick, maybe for those, uh, wa you know, those viewers, those listeners that aren't aware is that when we say Office 2016, office 2019, we're talking that traditional kind of legacy on-premises version of office that has nothing to do with Microsoft 365.

And so once that October 14th date hits, there's not gonna be any more security patches. And so if I'm a threat actor, you know, it's probably not that difficult for me to determine, you know, organizations that may have this particular version of office within their ecosystem. And that just gives me one additional place that I can target with a higher degree of success than, you know, maybe some of my more traditional avenues because I know that you're using now out of date version of office.

And again, once there's no security updates, uh, happening, you know, those holes aren't gonna be patched anymore. So, and I, I forget what the statistic is, exam, uh, exactly. But Office is one of those suites that traditionally has quite a few, uh, security bugs, uh, on the regular that have been patched in Microsoft Patch Tuesdays over the years.

So I don't think it would take very long before we would see some sort of security incident as a result of, you know, an organization running old versions of office. Will there be organizations that are still delivering for fee patches and things that I can use? Or will it just be no patches whatsoever?

Well, Microsoft has never been an organization to turn down money, right? So, uh, you know, it, it's funny working with them over the years. I sometimes, you know, they're kind of nebulous onto whether they're gonna continue to patch an application even in a extra paid capacity after the official end of support.

Usually it turns out that they end up doing it for a substantial fee because money talks, right? And if I'm a, you know, a 5,000 head organization and I wanna pay a Microsoft a hefty fee to continue to support my outdated version of office, they may in fact do that. That said, if Microsoft decides not to do that, which I, I think is possible, at least for Office 2016, because it's been out for almost 10 years now, right?

Uh, there certainly are, you know, third party like aftermarkets vendors, I guess I would say that have been known to provide patching for outdated versions of software. Now, uh, personally I have not seen any specifically saying that they're going to do that for office 2016 or 2019, but whenever there's a business need in the markets, the market usually provides, right? So it wouldn't surprise me to see that, uh, that happen in the, in the marketplace In a lot of organizations, there's still a split between the IT folks who manage systems and the cybersecurity folks who try to protect them.

Are the cybersecurity folks aware of this issue? Because, you know, it could be that the IT folks are just making decisions and cybersecurity folks gonna wake up one morning and go, excuse me, come again. What happened?

Well, I would imagine most organizations that have dedicated, uh, security folks on staff, they're probably very well aware of this. 'cause as a security expert, you know, if I was working for a a, an organization where I was responsible for their security posture, one of the first things I would do is create an inventory of all the software that's in use in the organization. And in my charts of said software, I'd have a column that here's the end of support date, right?

And there'd be alarm bells going off on my calendar that would, you know, keep me apprised of that. So I have to imagine that they are aware of it. Uh, if they're not, well, yeah, they could potentially be walking into a, an issue there, right?

But I think most security professionals are aware of this issue and are hopefully urging their, uh, you know, operations teams and infrastructure teams to, you know, get their house in order before that that date hits. Right. Are there smaller organizations that might be more at risk?

'cause you know, I go visit, say my local dentist sometimes, you know, they're still running stuff from some era of IT that I've long since forgotten about. Oh, for sure. I, I'm sure there are small organizations that are out there, uh, still running old versions of office.

And I, you know, I spent, before my tenure here at Hornet Security, I spent 10, 12 years, uh, in the managed service provider space, you know, in the trenches, right? And I mean, I can't tell you how many times I'd walk into a, a new customer or peck even an existing customer, or you'd walk in and, you know, you're replacing a new workstation or something. Come to find out they're using a version of office that's four revisions old or something like that.

They have no idea where the installer is for it, and they have no idea where their license key is. So I, it's absolutely possible that there'll be some SMBs out there and, uh, they're gonna, they're gonna be at risk too, right? Because they, they're less likely to upgrade than your enterprises are.

And they don't have the resources on staff to keep their eye on that ball either, right? They don't have a security person on staff, you know, saying, Hey, we need to update right now. That said, I will say Microsoft has done a good job of kind of nudging the SMB space in the mid-market, specifically towards Microsoft 365.

Um, they have made it uh, very financially advantageous for those types of organizations to move to M 365 because M 365 includes those security updates. They just perpetually get updates because it's a subscription based service, right? But there will be those organizations that, again, are still running old out of date versions of, of Microsoft Office.

Unfortunately, As we kind of ponder all this for a minute, do you think that, um, this is also gonna get tied up with an effort by Microsoft to get people to move from Windows 10 to Windows 11, and this is all part and parcel of that motion as well, because it seems like Microsoft is trying to get everybody on more current systems that, well, A, it makes it easier for them to support, but b, are actually more secure. You know, it's, uh, interesting that you mentioned that because I believe it's in October. There's a magic date for, for Windows 10 coming up in October two, I believe, where that's not officially supported anymore.

And to your question, I think indirectly Microsoft will be trying to resolve this issue through that upgrade to Windows 11. Because when you install Windows 11, it already has some soft hooks, I guess I would say into M 365. It makes it very easy for, uh, end users and especially the small businesses that we've been talking about to adopt M 365.

'cause it's presented to them right there within Windows 11, right? And so I think indirectly that's gonna be Microsoft trying to get people off of old versions of office. But, uh, of course we have the, you know, the LTSC 2021 and 2024 versions of office, that's like your perpetual, uh, you know, local version, uh, modern version of Office 2016 and 2019, which are going end of life.

Those are kind of that special case where I mentioned at the beginning of our, our talk here where if, you know, I'm a, an organization that for some reason Microsoft 365 is not an option for me, those two options, uh, L office, LTSC 2021 and 2024, those are the two places Microsoft would like me to go. Now that said, uh, if you go to 2021, you're only buying yourself about another year because 2021 is gonna go out of support, you know, not too long from now. So 2024 is probably where you wanna end up if you still wanna run that, that perpetual on-premises version of office, right?

Mm-hmm. Do we still have a problem with staying current with software? And I know historically we've all kind of tried to sweat assets and people are running older versions of systems and they're not willing to maybe upgrade as quickly as they might, but it feels like the cybersecurity risk has become much more profound.

And is the risk now greater than running the older systems? Because some people feel like, well, if I upgrade, I'll break my system. I'll have to migrate all my data.

And there's justs too much work for that. But I wonder if we're overcoming that inertia yet. You know, I'll preface this answer with the simple statement that if you can keep your software up to date, you should, I mean, full stop.

I mean, that's the, the number one thing you can do to, in terms of your software, um, attack surface to keep yourself safe is to keep your software up to date. Right? Now, realistically, you know, those of us that have been in the industry for a long time, we know that's not always a hundred percent realistic.

I'll give you a tangible example. Uh, I, many years ago, I used to support a, uh, a large, uh, manufacturing organization. And one of their buildings, they had I think three very, very large, uh, CNC machines that were used to cut steel for raw parts.

Now, these machines ran some ancient version of the software that was only supported on Windows 95. Now, this was 10 years ago, 95 was long outta Windows, 95 was, was long outta support then. And so what it comes down to as it professionals, as security professionals, it's measuring the risk.

I I think a lot of teams don't do a great job of measuring risk for certain situations In that particular example, what's the risk? Okay, we have a vastly outdated version of the operating system here. Uh, the, you know, the company, the manufacturer can't easily upgrade that software without significant financial impact.

'cause it was gonna be to the tunes of millions of dollars because they had to, they would've had to replace the CNC machines too. And so that's a situation where a software upgrade is not quick and easy. So in that case, what mitigations can I put into place?

And basically we completely isolated those machines from the network to mitigate that risk, right? And so, you know, it all comes down to, to risk management assessing that risk. But like I said, going back to the beginning of my answer here, you should be keeping your stuff up to date as much as humanly possible.

And then in this context, is it really worth the trouble at trying to upgrade to a new version of something like office, um, using existing systems that weren't designed to run that. So they don't usually have the memory or the processing capability to create a fabulous experience, or should I just go out and get new systems? You know, that's one of the million dollar questions when it comes to asset management for IT organizations, right?

And, uh, you know, office 2016 and 2019 have been out and in use long enough to where I would think for the most part, uh, many organizations have gone through a, uh, a hardware refresh, uh, somewhere in that timeframe. If they haven't and they've, you know, still running the same machines that they were back when they installed office 2016 years ago, they might be in a situation where it makes sense to, to upgrade the hardware. Uh, but of course then you're adding additional work to your plate of migrating to, you know, office LTSC 2021 or 2024 or M 365, right?

There is that additional, um, chunk of work involved with replacing out the hardware and refreshing it. But I think for most organizations, uh, most of their systems should already be in a place where, you know, they, they should be able to handle the newer versions of office. That said, those organizations that can upgrade to, uh, Microsoft 365 instead Microsoft 365, can largely be run inside of a web browser.

So if, you know, you're an organization that has some older systems out there and there's not a strong reason for you to keep, you know, your office suite on strictly on premises M 365 might be a viable option for you. 'cause really all you need is a web browser, um, word, Excel, PowerPoint, outlook. They can all be run simply in a web browser with, uh, minimal resources.

So that might be an option as well. Alright, What's your best advice to folks then as they kinda look at all this? Or conversely, what's that one thing that makes you shake your head and go, folks, we need to be a little bit smarter than that.

Yeah, so, you know, it's tough to give any, you know, magic silver bullet answer to something like this because every organization's needs are different, right? And so I guess kind of what I would suggest people do is if Microsoft 365 is an option for you, I would strongly suggest you look at that first, because you're gonna constantly get security updates. You're gonna get new features as they come out.

Uh, you get all the, the benefits of the M 365 suite, not just on the, uh, you know, the, uh, the office suite side of things, but you also get things like SharePoint and OneDrive and Microsoft Teams and all that other great stuff that's involved with 365. However, if you're in an organization that's, you know, like I mentioned earlier, if you have to adhere to compliance regulations that prevent you from using M 365, or I've seen some organizations that also just have a, um, you know, a general, I guess I would say mantra position that they don't wanna use the cloud. Uh, you know, that's when you're gonna wanna look at, um, uh, Microsoft Office LTSC, um, 2021 or 2024.

Like I said, look at 2024. Uh, you know, for those people that are ultra conservative and wanna be like super, super safe, you, I guess you could look at 2021, but like I said, you're only buying yourself about a year. So, uh, it makes sense to look at 2024 in that particular case.

All right, folks, there's an old saying about being penny wise and a pound foolish. I think it clearly applies here. Andy, thanks for being on the show.

Yeah, For sure. Appreciate it. And back to you guys in the studio.

Hey everyone, it's Shimmy and welcome to another Shimmy says, you know, I wanted to talk this week about something kind of near and dear to me, and that is DevOps. com. com in March of 2014.

Been working on it a while before that. And I mean, hey, timing is everything, right? I, I was fortunate and lucky enough, and it's better to be lucky than smart sometimes.

I was fortunate and lucky enough to be in the right place as DevOps ascended and became ascended. And, you know, over the last two years, we've, we've heard, you know, first there was that whole DevOps is dead thing. They get platform engineering oxygen, but that was pure marketing.

But we've seen, let, let's say less of a, less of a emphasis on DevOps, less attention on DevOps. It, it almost is just there. And that's not necessarily a bad thing for technology because I think it shows a level of maturity where people don't necessarily question it.

No, you don't hear what is DevOps does. DevOps works. Yes, it works.

And a lot of the DevOps companies, you know, call themselves DevSecOps companies, because that works too. One of the things though that, you know, gets me going is so much of the chitter chat, or so much of the oxygen getting consumed out in the social media sphere is by analysts who, you know, purport to cover the space. And I'm not saying they're dumb or bad or anything like that, I just question their view of the space.

And, you know, recently, I've had a chance to run into that twice in the last couple days. Before I talk about it though, I do wanna highlight a story that does get DevOps, right? com, and it's about my good friend John Willis.

You know, one of the greatest gifts that DevOps has given me is the chance to meet the people who really pioneered the OGs of DevOps, if you will. And other than Patrick Dubar himself, couldn't think of a bigger OG than my friend John. He, he really did so much to lift DevOps up.

com, that it's not just about the tools, it's about the people. It's about the culture. It's important.

It's not lip service if you don't pay attention to the people in culture. You don't really have DevOps. You just have DevOps tools.

And like any other tool is, it's just a tool. Um, so, you know, go John. And as I've said in the, in my article in many ways, John is still the true north, the north star for DevOps.

He hasn't lost sight, you know, him and Damon brought us cams and everything else. And so much of that fundamental foundational work around DevOps and DevSecOps is still true today. It, it's true whether you're a platform engineer or an AI engineer or SRE or anything else, DevOps is there.

People are using it, and it works with all of these things. Now, on the other hand though, I saw two analysts this week put out stuff on my, it showed up on my LinkedIn feed. One, one was a Gartner analyst who, who he's calling it like he's a doctor in the emergency room and the patient died that, uh, Schiff left is dead.

We should all be so dead. Schiff left has succeeded. Shift left.

If you were here when we started this thing you'd know is the fact that we're gonna start looking at security further left in the production pipeline. Now, what I think he really meant, and he just in typical Gartner fashion, maybe is trying to grab a headline or some click, is that developers are not security people. Newsflash, stop the presses.

We've been saying that for three years already. We can expect security people, the, I mean, excuse me, we can't expect developers to become security people. They care about the quality of their code.

No one wants to deliver insecure code. And what we found is if security companies make security tools for developers, it doesn't go over so well. And I think many companies have found that, and they've pivoted and, and come back to it and done it, right?

Companies like ny, for instance, right, are really, have, have fine tuned that really well, but moving security further into the development pipeline is not dead and it's not a bad thing. And when you get a firm like Gartner coming out and saying this, well, I gotta tell you the truth. I question, I question how much they know about DevOps.

And if you look at the history of DevOps, Garner's never been one of the leading lights in the DevOps space anyway. So yeah, developers aren't security people, but the rumors of Schiff left's death are very premature. And I, I think we'd be better off without that kind of kind of click datey stuff.

Another analyst, and, and it's actually someone who I respect from IDC. I'm not gonna mention names, but he came out with a DevOps DevSecOps market report just this week, I believe. 4%.

I, I'll be honest, I think that's a little conservative, but all right. I, I'm not, you know, I don't consider myself an analyst, though I did stay in the Holiday Inn Express last night. 6 billion market projection.

What gets me is, here's his market analysis, here's the companies he covers as the leading lights in the DevOps tool space. And I just wanna make sure I get this right. Leading vendors in the DevOps software tools market.

Yes. And here they are. Microsoft, Atlassian, Datadog, Dynatrace, Broadcom, IBM, and New Relic.

All good companies. Are they the leading lights in the DevOps space? IBM hasn't mentioned DevOps in about three years.

They, they've moved away from DevOps, they're into other things. Datadog's a leader in observability, Atlassian, JIRA. Absolutely.

I could see that being in there. Not Dynatrace is a decent company. New Relic, they've had their ups and downs, but let's call 'em what they are.

An observability company. Microsoft, I'm not sure if he's talking about Microsoft Azure, DevOps or GitHub, which is part of Microsoft. Or maybe he doesn't consider DevOps a GitHub, a DevOps tool, though it's clearly A-C-I-C-D tool.

com and Cloud native now and platform engineering. How do you do a report? And granted he has rest of market represents 55% of the report.

How do you even put out this report without mentioning the leading DevOps companies? So to me, this is e either Tone deaf, maybe these are the companies that subscribe to them for research, I don't know. com.

Now I know what you're gonna say, shimmy. You guys are part of future, you're an analyst firm too. Yeah, we are.

The guy who covers DevOps here is a guy. You've probably seen him if you've been around Techstrong, his name's Mitch Ashley. You ask Mitchell who the leading DevOps companies are, and he's gonna tell you, and I guarantee that he will name most of these, but he's gonna name most of the ones I named too.

And then some. I I, I don't think these analysts do US justice or themselves justice when they put out reports or, or statements like Shift Left is dead, Uh, Et cetera. You know what?

DevOps isn't going anywhere. DevOps is well settled. You know, it, it's just part of the part of the way we do things.

But I expect more in a world of AI where these guys can go out and do research pretty easily to have better, better information, better conclusions. As a matter of fact, with ai, I don't know how much longer we're gonna be relying on these kinds of reports. 'cause I think AI could pull this data itself themselves.

We spoke about futurum Signal, I think it was last week or the week before it, you know, this is the AI up to second reports. Instead of getting a report here in August about the 2024 DevOps market, how do we as a, as an industry function on reports that talk about things that happened at best nine months ago, eight months ago? It's beyond me.

Anyway, here's a little advice you want to keep up on the DevOps market. com, watch our DevOps, uh, unbound control alt deploys, Textron gang, cloud native. Now we cover DevOps every day.

You wanna watch stay on top of DevOps, follow what Mitchell is, is doing over at Mitch. Ashley's doing over at fu because if this is the best they got, it's not good enough guys. I'm sorry.

You know what, Shimmy's never short of an opinion. I call him as I say him his garbage. We're outta here.

Have a great week. Shimmy Says, Welcome To six five Summit. We are in our sixth year and we are talking about all elements of AI from enterprise SaaS to infrastructure to security and everything in between.

Yeah, it's been a really great event, pat. This is always our, our big moment. The who's who some of the best thinkers in the industry couldn't be more excited.

I mean, we are literally now what in the third sort of big year of waves of ai and we are unleashing this year. And what's been really important about this year versus other years is we've really gone from theory to application. And I think this is what the world needs.

We've kind of talked about what might happen. That's right. Now it's time to talk about what is happening.

That's right. And Daniel, uh, a lot of the narratives around, uh, ai, right? They center on, you know, the infrastructure server, storage, networking.

Uh, one of the things though that just doesn't get enough discussion, quite frankly, uh, is security, right? The current security stacks are not made from the age of a age of ai. And I think that needs a double click.

And I can't imagine a better guy to go through this than GTU Patel, uh, CPO at Cisco. Great to see you. Welcome back to the show To see you as well.

And Congrat six years on, that's a congratulations. You've been Doing this for a minute. I mean, you're a regular on our show.

I, I'm, I'm honored. Yeah. Thank you Val.

I'm So glad to have you at the event. Um, you know, not only chief product officer, uh, newly minted president Yes. Jobs to anyone these days.

Man, my gosh. Tell you what, If I hadn't talked to you so many times, I might have bit on that one. Well, I mean, it's great to, I mean, listen, it, there's a different energy now with products and you know, I, I feel like sometimes even though you're this big company at Cisco, like I'm talking to a startup, and I think that is really, that's the goal important.

Like having the, the psyche of a startup and the resources of a large company are really a killer combination. We, You know, one of the things that we actually did was, uh, that was one of our mantras in the team to make sure that we operate at the speed of a startup at the scale of a large company. And, you know, if we could do that right, and the, the kind of recruiting you need to do is slightly different as well.

So you guys met dj, um, as well. Uh, he's the guy that runs our AI business. And um, you know, he came, he was the CEO of, uh, one of the companies we acquired.

And we've got probably like, you know, eight, nine CEOs of different companies. We required like tj who are running different parts of our businesses. And then we've got people that know how to navigate Cisco.

And the combination of those two, if you pivot too much one side or the other, it doesn't work. But if you get the two of them combined, magic starts to happen. That's incredible.

That kind of operating leverage. But also, but it is also, and, and you and I know this 'cause we evaluated money or something comes, getting that right is really, really hard. Let's talk about cybersecurity, a little OG too.

This has been something for you that's been a, been a passion project to, to not only reinvigorate, but really start to scale that part of Cisco's business. You had a big moment at AI Summit. You launched AI defense.

So this is at the epicenter of everything. But you talk a lot about how difficult this is. You know, AI is actually changing the entire calculus for cybersecurity.

Share a little bit about, you know, kind of how companies are managing, how you recommend a company sort of manage this inflection and this moment and how to deal with the risks cybersecurity's creating. Alright, Firstly, if I were to think of it, the, the use of AI in cybersecurity trails that in other industries, and that's, that's actually a, um, um, a function of a couple things. One is the efficacy with the use of AI of cybersecurity is low right now.

And we'll talk about why that's low and what we can do to make it high. And the cost is too high if you use ai. And so those two things have to get fixed.

And then the second thing is, this is a tremendous talent shortage in the industry right now. And so we have to make sure that, and this is where AI is a huge opportunity, because there's no downside. You get 4 million jobs that go on field every year.

You actually have AI agents that can be augmented to your workforce. It's just goodness. You know, like there's, there's a lot of industries whether that can trade off.

Like, you know, is this good or bad? If I have AI over here, there's literally no downside. You have to have ai, there's no other way around it.

And so I, I feel like, um, over the course of the past six months, uh, the inflection point has really started to come with ai where you're moving from this world of, you know, chat bots answering intelligent questions to agents, going out, kinda getting their jobs done, um, in a fully autonomous fashion. And we now need to make sure that we completely change not just the cyber defenses. So if you have, there's two kinds of things.

One is using AI for cyber defense, and the second one is securing AI itself, right? And on the securing ai, it's products like AI defense, it's say, Hey, I've got this unpredictable non-deterministic model that is now a fixed part of my architecture on top of which applications are getting built. That and my applications I'm building as an enterprise need to be predictable applications.

So it just doesn't make any sense to build it on an unpredictable model unless I can get that unpredictable model with guard rails. Otherwise, it's like very, you know, it's very hard for a company to bite the bullet. So I feel like what used to be a few years ago, a competing alternative, like if, do you wanna be productive or do you wanna be sec uh, secure, those were competing alternatives, right?

In ai, if you're not secure, you can't be productive. You can't drive adoption because if people don't trust the system, they're never gonna use it. And so we have to completely change how securing AI is done.

The way that we do securing AI is it's a three part process. Get full visibility in what's happening in your state. Number two, get complete validation of how these models actually work.

Do you want 'em to work in a certain way? Are they not working the way that you want 'em to work? Can you jailbreak the models and when you can, can you put runtime enforcement guardrails on your applications that you're building so that you have one common substrate of security across every model, every application, every agent that you build.

I think that's a hard thing to do. I don't think people know how to do that. Um, you know, at at scale easily right now.

And the technologies are just starting to come about. Like we've got so much traction with AI defense because of that reason. Yeah.

I get this question, uh, a lot, which is about, uh, why security AI I is so hard. I mean, you talked about the non-deterministic, uh, and that would be hard, uh, to protect, but also, you know, is is it the data estate change? Is it the endpoint pervasiveness?

Uh, what is it that makes secure AI so difficult? Actually, the, the thing that's difficult more is the models that we build these systems on tend to be generic models. And I always tell people like, don't, don't use the model that you used to write poetry or do a do a pizza recipe for cybersecurity because those might have different kind of data sets you wanna train 'em on.

And so the reason it's so hard is because these models have been generic. We are now entering into a world where we can actually have far more specialized models. In fact, Sam Altman had a great line that he talked about in one of the, you know, kind of conferences he was at.

He's like, the future models are gonna be, I mean, you, you're always gonna have large models, right? That's always gonna be the case, but there's gonna be a different class of models which are small models, trillion token context window, and connect it to everything. 'cause today the models are acting as databases.

And if you can connect the models to everything, then you still have the efficacy of the model being small. And what we did, um, at RSA was we announced a, um, security model that was very specific and bespoke for security that we open sourced. And then since then we've quantized that model.

So what used to be something that was so small that it could run on one a 100 GPU, we can now run it on a laptop on A CPU. Yeah. And that imagine the cost curve differential that happens with that.

So once you've got a model, um, that is, uh, efficient, high amount of efficacy, it beats a 70 billion parameter model. It is as good as a 70 billion parameter model, um, with, uh, with a fraction of the footprint, right? If you have that and then you start building applications on top of it, you get a very different kind of outcome from it.

Sure, Yeah. We've, we've actually done a lot of valuation in our, in our labs G two, where we've looked at these smaller language models. Yeah.

And they can be extremely eff and um, they're also very efficient. So yes, much less power. And we know in this era, because you and I, uh, actually all three of us had a conversation about three big constraints, right?

And I mean, two of 'em are being powered network and, and the others in network, everyone understands compute and so the other two don't get talked about as much. Yeah, that's right. That's Right.

But um, you know, getting these more efficient, getting them smaller, moving them to the right sizes is going to be really important. Now, you mentioned something in the beginning of the show, you know, where you sort of talked about why, um, you know, security trails. Another thing that is historically driven, the trailing of cybersecurity has been just how it's prioritized.

It was for the longest time, like, you know, the board would be like, well, what's the least we can spend? So another big problem has been kind of the spend for in the AI era, though everything's happening too fast. If you're a company, you have to put security at almost the same level as AI in terms of your prior, it Could be a business ending event if not done Single handedly.

Yeah. And, uh, you know, There's not too many things a CEO gets fired for, but a security breach is one that they can Absolutely. And then you add autonomous agents that are gonna have, they're basically employees.

That's right. Putting Pos, sending wiring information out to people, helping with airlines. I mean, it could get really serious.

In fact, the identity associated with, um, agents and having zero trust, not just for users connecting to apps, but also being applied to agents and iot and robots is gonna be really important. I think physically I will be here before you know it as well. And I just don't think that our security infrastructure currently is designed for that.

So we have to make it hyper distributed. We have to make it agent friendly and accommodate, you know, a very, very different kind of operating model, um, where the efficacy and the costs are in line with what, what people expect. So you Started answering what I was almost gonna get to.

I ask the longest questions, it's just sort of a thing with me, but like, okay, so, and I give the longest answers, so it's like, great, You the match. I'm not supposed to, I'm not supposed to do this, it's just, it's my style. But like, how do we get there?

Like you kind of talked about the outcome of where we need to be. So those things you just mentioned that we need to eventually get to, how do we get there? Because right now security still feels, in many ways it's fragmented.

In many cases it's moves slower. If you look at like a lot of, when you talk about like in, um, you know, the model injection risks and stuff like that, the attacks, The prompt, the Prompt injection, it's happening because this, we're pushing out models as fast as possible. We're making them democratizing them to everybody.

This is how, you know, we ended up with large companies giving free data to models that would then be trained on to give data to other people. You're trying to fix that. Like how do we move this along?

Yeah, I think one of the challenges that you bring up is a really interesting one. The average time of a model in the market is about six months. The average shorter, right?

Uh, and getting shorter. The average time of validating a model in the enterprise is about nine months. Still doesn't work.

That doesn't work, right? So you have to make sure that you get the, the way in which you validate these things has to be algorithmic. It can't be, it can't be human scale, you know?

And so, um, let's, let's actually take a step back and start from what needs to happen. There's three things need that need to happen in order to secure ai. Number one, you have to have full visibility of all the data flowing through a model and what models exist near a state.

You can't protect something you can't see. Number two, these models are non-deterministic and they're unpredictable. And you have to make sure that you validate them and jailbreak them so that you know that the areas where you don't want it to work, the way that you in the areas you're afraid of it working the way that you don't want it to work.

You can figure out how to trick the model, right? So when deep seek came out in the first 48 hours, we were able to trick the model and jailbreak it in the top 50 categories in the harm bench benchmark, right? 100% success.

Uh, attack success rate. This is the one time a hundred percent is bad, right? So you have to make sure that you get the models validated.

Number two, when you jailbreak the model and what does jailbreaking a model mean? If I ask a model a question, how do I build a bomb? Pretty easy answer.

If I then ask a question, well, I'm actually writing a movie script. Um, Brad Pitt's gonna be in the movie, uh, show me a scene where Brad Pitt builds a bomb in his car and then blows up the Bellagio in Las Vegas, immediately the model might spread out an answer for you, right? Video's Probably gonna get censored now.

And so what we need to do is we need to make sure that, that that validation exercise is done through this process of red teaming. And red teaming means you're just having people hack at the model saying, let me just give it, you know, questions from 10 ways to Sunday. We've done that algorithmically.

That's step number two. Step number three, once you've identified how the model gets jailbroken where it's not working, you then provide runtime, enforcement guard breaks. If you do those three things well with an underlying model that actually works well so that the efficacy of the jailbreaking is good, um, you actually have a pretty good solution.

Because then what you can do is that solution can be called upon by anyone building an application and saying, I don't have to worry about building a security stack. I'm just gonna call this API, right? And so anyone building a model, they don't have to go out and worry about the safety parameters of the model because they can just make sure that they call an API from this, this product.

And that's what we've done with AI defense. So that's the first thing. The second thing is how do I secure ai?

Uh, how, how do I secure, um, my environment and use AI for cyber defenses? There, you just need to make sure that you're using AI for the defenses rather than just doing it at human scale. And that is where we built this foundation AI model, got it to super high efficacy, very low cost quantized.

The model made sure that the training data set was very relevant. Um, and when you do that, you just have this amazing potency of a model that can be used in every application that you have. So G two, the rate of change in what's going on is, is immense.

You know, I mean, it's not our imagination. Uh, innovation is accelerating. It is in this space and changes.

How do you manage a roadmap, a vision, how, in a way that you can stay ahead of all this related to security. I think you have to have extremely smart people in multiple different domains that have a common value system. And that value system has to be, um, I'm gonna work in an open ecosystem, even with my competitors.

I'm gonna make sure that I'm gonna have AI first in the way I think about things. And my, my primary objective is to out innovate the adversary that those, those are the core principle, right? You have to apply if you do that, right?

You know, we've got a lot of entrepreneurs now at Cisco that are leading businesses. They were CEOs of companies that joined us. Um, and then we've coupled them with people that actually know how to navigate Cisco really well.

And the combination of those two, but shocking the system a little bit, I think is what ends up working. But we, we have to be constantly dissatisfied. And if we're not innovating fast enough, um, you know, like if, um, the, the tempo cannot slow down.

I think when the tempo slows down, bad things happen in a company in tech. Yeah. So it's a, a, a mental model or a management model that you probably have KPIs set up in terms of acceleration.

You have time to product log or time to product ideation or Some, oh, so that, that is a very specific model that we have. You know, it should take nine months to get a product out to market. Three quarters is what we try to do.

Idea to product, to market with ai. Hopefully that actually goes down, right? Um, once you're in market, uh, the first thing we focus on is obsess about getting to product market fit, which means is the product working the way that it's supposed to work?

Solving a problem that we thought it was gonna solve, where the customer says, if you take this product away after 30 days, my life is gonna get meaningfully degraded. We should have at least 40 to 50% of the customers that feel that way. Otherwise you haven't achieved product market fit.

Number two, get to go to market fit, which means you have a repeatable opportunity creation motion. I'm gonna keep creating opportunities with the same titles and the, and same class of companies over and over again. If I don't, then I'm just selling to friends and family that's not scalable, right?

And number three, invest in scale. Most companies have not gotten to product market fit, skip, go to market fit and start investing in scale. Bad idea.

I think you have to do it sequentially. I like the constant disappointment 'cause anybody knows me knows that that is how I live my life. But I think that, I mean, it's a s****y way to live life, but it's actually a really good, uh, is, it's a really good thing for Business.

I think it's a pretty strong characteristic of, of successful business people. Like I always joke at the end of the month, you have a great month or a great quarter, and then instantaneously the first day of the next quarter is, what Have you done for me lately? What have you done for me lately?

And in, in cybersecurity? It's very much the same thing. Gtu Patel, I wanna thank you so much for opening up our cybersecurity track here at the six five Summit.

Thank you for having me. Thank you everybody for being here with us. Stick with us for the six five Summit.

com/summit. If you wanna check out any of the other tracks, appreciate you tuning in. Stick with us more.

Coming to you soon.