Governing AI Agents as First-Class Identities

Hey everyone. Welcome back here to Techstrong TV. I'm happy to introduce you to my next guest.

I was telling him offline, we've dealt with his company for a long time, but I've never had him here on Techstrong TV. So we're really happy to welcome Ravi Sawhney. Ravi is the CISO, Chief Information Security Officer, at Smartsheet.

And Ravi, welcome to Techstrong TV. It's great to have you on here. Yeah.

Same here, Alan. Thanks for the time and appreciate the interview. It's my pleasure.

So Ravi, just thinking where to start here. People want to know who they're talking to, or actually who they're listening to in this case. But Ravi, give them a sense of how did you come to be the CISO over at Smartsheet?

What's your journey been like? Yeah. Thank you, Alan.

Yes, I, as the CISO, joined Smartsheet about nine months back. I was, prior to Smartsheet, spent about 15 years at a healthcare software company called FX. And then before that, in various product development roles.

My journey as CISO has been through the trajectory of software development. So, building software at scale, enterprise scale. And prior to my arrival at Smartsheet, I was also a customer.

So I was familiar with Smartsheet as a user, as an enterprise software company. And fortunate to be here. And started that lens of the product development mindset, and engineering mindset and taking that from a CISO perspective gives me a unique advantage from that perspective as a software company.

Excellent. I love it. Ravi, as I mentioned, I've known about Smartsheet for a long time.

We've worked with Smartsheet for a long time. And I think most of our audience has probably heard of Smartsheet. It's been around.

But there are probably people who haven't though, to be fair. And so if you don't mind, for those who are not familiar with Smartsheet, give them a, I don't know, not the elevator pitch, but tell them what they should know about Smartsheet. Yeah, no.

So Smartsheet is, we are an intelligent work management platform that really unites the people, the data, and the AI to really turn your strategy into sort of measurable impact. We are deployed in 85% of Fortune 500 companies and the projects, over 3 million active projects globally in our platform. So what's unique about Smartsheet is the notion of you look at an enterprise stack, right?

You've got tasks, you've got AI that is executing some of the workloads that are there. And Smartsheet's platform allows the enterprise to take those AI interactions, whether it's your data, whether it's the workflows that you have in it, and custom assets that you may have, and really pull together a controlled environment or layer that is designed for transparency and compliance. So that allows for much more, from a work management platform perspective, this unique blend of taking tasks, timelines, risk, and dependencies, and having a very native integration across all of the supporting infrastructure that we have.

I love it. That was an excellent job. Good way of describing it, and I think that makes sense to our audience.

So Ravi, if we now can, I wanted to pivot into what I wanted to talk to you about today. We're living in, I call it historic times. Right?

com era. I started a company, I sold it, I took another company, went and helped them go public. So I went through the ups and the downs.

And by the way, rode that IPO up and all the way back down into bankruptcy. That's how crazy that was, right? So I've seen the ups and the downs of these waves.

This one feels even bigger, different, but it has that same thing I loved about when the internet first went commercial was you weren't just changing technology. It was like changing society, was changing humanity in some level. This has that same sort of feel, this whole AI thing.

But the tip of the spear is technology, right? Whether we're talking about developing and cogeneration or security and cyber, because the bad guys have this too. It feels like we're in very unsettled ground, right?

The ground's kind of shifting day by day, almost hour by hour. And so what's a poor CISO to do? Right?

Because you've got a board who's saying, "More AI, more AI, more AI," right? And what about security? What's my risk?

At the same time, you know very well as a CISO what we're up against here, right? It's no longer about just finding vulnerabilities. We can almost find vulnerabilities at will.

We got to fix them and defend against them. com or the advent of the cloud or anything else, what can history teach us here, Ravi, in these- I know ... you know, to help?

I think definitely lots of lessons from history. I'll take a sort of a an analogy slightly outside of cyber, then I'll connect it back to cyber as well. So, you can take the incident that happened in the Pearl Harbor as an example.

1941, the signals that were there in place, there were in fact 14 different signals that the US had intercepted from Japanese intelligence, pointing to this imminent attack. The data existed. What failed was the system for connecting all the dots and acting on them.

And the AI peril is very analogous to that. When you think about enterprises today, you have more security data than we've ever had. Whether it's logs, you've got model outputs, you've got agents actively tracking your signals and so forth.

Yet AI-related incidents keep escalating. It's not because the tools don't work or you don't have necessarily the signals, but because, frankly, organizations lack that governance layer to correlate what AI is doing, assign accountability when something goes wrong, and act before that danger propagates. So, you sort of saw this also back in the late 2019, 2020, the large SolarWinds one as well, where we had the Russian state actors inserted a backdoor into trusted software update that happened in '19.

It sat undetected inside 18,000 organizations, including Microsoft and the US Treasury, and for 14 months. And so the telemetry definitely existed. The security tools were there, they were logging and all of that same stuff.

But again, it goes back to that notion of governance and that accountability layer that we've had and ensuring that capability, or what I call sort of in my mind, capability without synthesis, and that was the gap that existed. So these are real historical things, and I'm listening to them, and they're making sense to me. And, Robbie, a theme or a thread through them is the hints are there.

It's almost like, how do I read the signal? Sometimes it's signal lost in noise, by pulling the true signal out from the noise. Sometimes it's you've got the signal, and you just don't recognize it, or you don't want to recognize it, and it goes from there.

Yeah. Here, so again, I'll ask you, what's a poor CISO to do? I think we see the signals.

The signals are we've got to get ready for AI scale. Everything we've built to this point seems built to human scale. Mm-hmm.

How much code my developer can-- How many lines of code can a developer do in a day? How many vulnerabilities can a researcher find scanning? Mm-hmm.

How much remediation can one person, patching or other, can one person? Our whole environment is built at human scale, and now we're saying, "Okay, clear the decks. " The velocity is exponentially faster.

The scale of what we can do, it's 24/7, 365 at machine speed, exponentially more. What's a CISO to do? And- Yeah, no ...

go ahead. Alan, I think, I would say that the first thing I will say is the basics are unchanged. We were, as an industry, betting on the time that it took to look at our attack vectors and the time it would take for a bad actor to get into your environment and do lateral movement and all that other fun stuff.

So, if you look at the trajectory, that exponential curve that has happened on a downward scale, where the time of exploit has significantly reduced to hours, in some cases, minutes, from what used to be months or days when a CVE was announced. So, if you see that exponential logarithmic scale coming down at the scale and the velocity we have, the fundamentals have not changed. We need to focus on building and scaling software, and with the right processes in place.

So thinking about secure software development lifecycle, ensuring that you have your vulnerability management practices in place. You've got the right, very basic things in place. When you're checking in code, you're checking in the right amount of your patching and all those basics we have talked about as an industry.

The gambling that we were doing around time is no longer as lax as we were in the past. So, I go back to, I think if you think about the frameworks that have been around for a very long time, around NIST or controls and shipping software that's associated with it, I would say that none of that has changed. If anything, the only thing we need to look at from a governance layer on the AI side is the metric framework and how you track AI agents.

We talked about AI as identities. Those are the same principles that we've talked about as human identities. Like how do you protect those?

So I think the fundamentals have not changed. If anything, the role of a CISO is to ensure that the basics that we've been trained for decades, you're increasing the velocity of ensuring those fundamentals in play. I agree.

And again, that's a history lesson there, too. You know what? Laws of physics don't change, Ravi, right?

Best practices usually don't change. They're grounded. They are best practices for a reason.

So the things, they're bedrocks that we count on. We just, as I said, we may have to do them faster. We may have to have more scale.

But those fundamentals are still there. Mm-hmm. What else, though?

Because the CISO, I've always, and as I said, been in security a long time. CISO's in a funny place. They're at that junction.

On one side, they've got to talk business to the rest of the executive team, to the CIOs- Yep ... to the board. But on the other hand, they've got to be technical enough to talk to their security teams about the bits and the bytes, the CVEs, and everything else involved in there.

What can a CISO do to stay on top of what seems like shifting ground? Yeah, I know. I think my first ask of any CISO is thinking about the governance layer, right?

And when you think about governance, there are three specific areas, right? For every AI agent that you have in your organization, you need to ensure that it has first-class identity, right? Named owner, defined permissions, do you have a documented life cycle?

Not this anonymous processes running in the background. And, the second thing is the blast radius boundaries, right? What is each agent authorized to touch?

What are the hard stops that you have at the edges? And that's extremely important from an AI perspective. And the third is the circ, what I call the circuit breaker, right?

You have an automated mechanism that halts execution when you have an agent that encounters something outside of the expectation parameters. And do you allow, and can you stop it in the right time before these actions propagate downstream? So I'll give you a little bit of an analogy, right?

You think about the Colonial Pipeline back in 2021. Yeah. We had the ransom group, DarkSide, that gained access to the pipeline through a network of single compromised VPN accounts, right?

There was no MFA in place. One credential, one entry point, and within hours, the largest fuel pipeline in the US, right, in the East Coast was shut down, right? 45% of East Coast fuel supply was gone in six days.

And that had major gas outages, and as people may remember, across multiple states. So, the guardrail that was missing was, in my mind, very elementary, right? And these are things that we would consider table stakes at this point in time.

Yet its absence had this catastrophic cascading effects. And so I think the principles that we think about AI agents around having the governance around what those permission structures are, treating it as a first-class identity, is absolutely paramount. And having those guardrails that I just outlined earlier makes sense.

I love it. Ravi, we only do these for 15 minutes. We're almost out of time here.

For people who want to get more information about Smartsheet, what's the website? com. Yep.

And if they want to follow you and your ramblings, writings, thoughts on the challenges we're facing and how we can deal with them, is there a blog, LinkedIn, any place they can go to? LinkedIn would be a great place to start. com/ravisoin, all one word.

I know Smartsheet also has published articles and so forth in Smartsheet that you can follow as well. And I definitely talk about security and privacy and all the fun stuff that goes around AI and agents that are there as well. So I encourage you to also follow Smartsheet.

This probably wasn't on your radar when you first got into security, right? That you're going to have to worry about AI. But that's the world we're in today, and we'll do the best we can, as we always do.

Yeah. And exciting times ahead, for sure. Absolutely.

Looking forward to it. Ravi, don't be a stranger. Come back on and keep us posted, okay?

Will do. Thank you so much for the time. My pleasure.

com. Check it out. We're going to take a break here on Techstrong TV.

We'll be back with more. We got a lot more coming your way.