The AI Security Regulatory Landscape: Government Leading The Way | RSAC Virtual 2024

Okay, well, good morning everyone. Happy RSA. Thanks for joining me.

It's a fantastic honor to open RSA with you all. Um, thank you for having me. Uh, I love this.

I love this particular session, and I'm excited to talk with you about policy. Um, you know, when you think about ai, uh, it, one of is, it really is one of the most fascinating domains of technology because we do have the regulatory landscape changing as we go. You know, in cybersecurity, we had years to think about, you know, the attacks and consider requirements and regulations, and then we adjusted as we go with supply chain and, you know, IOT and ot.

Here we are really seeing this moment where the same time governments are thinking about attacks, the same time governments are thinking about how AI is being used in the context of society and economy. Technology is, you know, evolving, attack surface is evolving, and we have the requirements rolling out. So fantastic day and time to talk about policy.

And, you know, my goal really here is to set the stage, uh, because all the attack vectors you're gonna see today, all the technologies, all the amazing, um, you know, solutions, uh, they're gonna be adopted in the market based on market needs that are really being shaped by, uh, regulatory requirements. So I wanna give you the context of what Washington, what Brussels, uh, is considering right now. Uh, and just, you know, really, um, uh, set stage on the requirements.

So, um, open policy, uh, is my company, co-founded by me and, and David z. Uh, we are the world first policy, intelligent and intelligence and engagement platform. We are basically aiming to bring government affairs value the same type of function that organizations like Microsoft and Google have emails to unicorns.

Um, and we've been running for a year. If you check out my RSA talks, just last year, I actually announced that I'm gonna be leaving Intel and I'm gonna be co-founding open policy. It's been a year, and, um, it's been fantastic to come back here with our customers, some of the world leading technology companies, companies like Hidden Layer that won the, say Sandbox last year, companies like Armies, uh, and, and we have more that are not even listed here.

So, uh, we are excited to bring the voice of entrepreneurs and innovation, uh, to Washington, to Brussels, and help those companies understand what is their future market by predicting what are gonna be the regulatory drivers that are gonna be applicable to the CISOs, to professionals like you, but also help them come together and engage. Um, one word about me, and I'll be fast. Um, so I come from this background.

Um, I'm formally the head of cybersecurity policy at Intel Corporation. So I manage the government relationships and policy for Intel. I also teach at uc, Berkeley.

Um, I have a PhD. The doctor is from the law. It's a JSD, uh, so four degrees, three of them in the law.

Too much I know. And fun fact. Um, you know, I, I came to the United States to Berkeley, and my first kind of coalition, my first exposure with PO for Policy was creating a coalition for hackers.

If you heard about the Safe Harbor movement or disclosed io, this whole idea of having protections for hackers in back bounty programs, um, it's, it kind of started for some research that I've done at Berkeley and a lot of other work of the community that ultimately got me headhunted by Intel. So I really love coalitions, and specifically I love the ability of the security community to come together and work with Washington. So that's a little bit about my background.

My co-founder is a second timer, CTO, and he is really, uh, running, um, you know, the development team and allowing us to use the cutting edge AI ourself, but to just to make policy more accessible. Um, you know, really we focus on digesting the avalanche of proposed regulations and connecting it to business roadmaps and products so we can prepare our customers and our ecosystem to what's coming. And at the core of, you know, what we're trying to achieve is this understanding that the future of security is pretty much dictated by, um, compliance, right?

By regulatory requirements. You do have the attack surface and then threats and the technology, and that's super important. But compliance requirements, the need to comply is a lot of what is kind of underlying CISO's decision making and where to spend money if you know what's coming, you can create better products, you can adjust your marketing if you can influence it, you know, you can really make sure that kind of your voice is out there together with, with the big players.

Uh, and in this world of avalanche of, um, requirements, we do need to reduce complexity. And this is where we're using technology and crowdsourcing to, to kind of, uh, parse through the noise. Um, today, you know, uh, this capability of government affairs that we think is one of the most important business development capabilities for organizations, uh, is really hosted with the large companies.

And, you know, those companies are great. They're doing a lot for society, and they have the access to lobbyists and trade associations, the government affairs professionals, uh, pretty much everyone else is not in the conversation. They don't see what's coming, or it's very hard to understand because there's a lot out there.

And we are aiming to be that rich, uh, to connect, uh, them so they can see what's coming, but also influence it. When we talk about entrepreneurs, we talk about regulation as the future of the market. You should know what's coming in the executive order on ai and what's gonna be the requirement for security.

So you can be ahead, prepare your infrastructure. If you're a company that is developing security AI solutions, you need to understand your customer needs. We are in the position that sometimes waiting for the CSO might be too late, because by the time the CSO knows what are the requirements, you are already behind you, were supposed to drive those innovations and develop them and understand the customer needs so you have enough time to prepare.

And basically looking at future, uh, requirements coming from government is a shortcut to that end. Um, and a lot of the, of the issues that we see is like, okay, there's so much going on. Like even in ai, any guesses, like how many deliverables and documents, regulations we have coming under the executive order on AI being developed right now, just for security, I'm gonna estimate about perhaps 25 just security, just AI security under the executive order that we saw from the Biden administration in the end of October.

So there's just a lot of documents, uh, and, and we help our customers focus. Uh, ultimately what we are is a one-stop shop, tech enabled shop for government affairs, value, it's intelligence, uh, documents, looking from everything coming from government, but sipping it with AI in a way that is connected to your business, but also allowing our companies to come together into coalitions, like mini trade associations and actually file comments and engage with government. Um, so they're, they can also influence, they can also impact not just see what's coming, but actually engage in the conversation.

And there is a lot of crowdsourcing that goes into that. Okay. Um, now we're gonna talk about, you know, all our, all everything that we've been trying to do for the year we're running sufficient to say that we've been in, in, you know, important forms, including on AI or our names, uh, was listed on the executive, uh, order.

I really wanna dive in into the regulatory landscape. Okay? So, uh, we are in a moment, uh, when it comes to ai, because the conversation about AI is not just impacting the requirements for all ais that are getting rolled out, including on security, it's reopening the way governments are looking at, uh, threat.

So it won't be surprising if the whole understanding of zero trust, for example, including foundational documents coming from government on how to apply zero trust architectures are gonna get reopened because we have ai, not just this code, but there's something that is considerably changing the front landscape. So this requires adjustments, adjustments from the main tool that we have in order to reduce risk to society, which is best practices and guidelines and requirements. And I really wanna double down on, you know, two major things that happened in the last couple of months.

Um, towards the end of last year, we had the release of the executive order of the Biden administration on ai, which is a very comprehensive action. Uh, it's considered, you know, some debate if it's, it's actually more impactful than the reactor. There are discussions on this, but it's considered the most impactful action the US has done on ai.

The nature of the requirements on AI day actually extend way beyond some of the privacy regulations that we have in states, in the United States. Um, this executive order is applicable mostly towards AI systems and software being used by government. But if you think about it, if your ai, you know, most of the, a lot of the commercial off the shelf code software is going to civilian agencies, right?

So it's a very, very broad scope. And if you thought, if you've heard about how many of you heard about the cybersecurity executive order, right? A lot of hands, you know, he'd deep impact, right?

Deep impact on industry, that's an, and even perhaps more, that's the impact we're gonna see from this executive order. And it's not gonna take years. You know, the, the cyber executive order is now finally getting into far like the federal requirements is are beating women, and we have the self decision form.

Uh, we already saw a memo last, um, last month coming from the White House saying, agencies, you have like six months, six months to start looking at AI inventory, data inventory, implementing those requirements. So this is not gonna take years until, uh, it's getting implemented. In fact, if you have a piece of software that is going to contractor that is somewhat somehow covered by the executive order, you should be thinking right now about your security measures, about your inventory, because those requirements are gonna be rolling down, right?

Rolling down the contractual, um, supply chain. That's the executive order. We'll talk briefly about the requirements there, then we have, of course, the UA Act.

Uh, who heard about that? Some hands. Okay, a little bit less.

Um, perhaps this is one of my biggest takeaways for you. Security, uh, requirements are being dictated not just by security regulations. So the UA Act, for example, has a lot of security requirements for ai, and you cannot be ignoring what's going on on the UA Act and just focus on things like the Cyber Resilience Act or the NS directive.

This nexus of risk between cyber privacy and ai. All the requirements are kind of bundled, especially in Europe. Uh, the U Act is the most comprehensive action on ai.

Uh, we call it the brussel effects, uh, with, it's like GDPR when Europe goes ahead and, you know, regulates with very robust requirements. And then because of the breadth of AI technologies that are coming into Europe, it's just impacts the whole market. Uh, they are, uh, it's risk based, meaning there are some applications of AI that are considered high risk, and for them, there are more requirements.

Uh, but the category of high risk AI is pretty broad, right? And can sweep in, you know, foundational models. Um, together those two developments are really shaping the core of the security requirements that are coming up.

So let's, let's dive in quickly. Uh, we are very, uh, fortunate to be engaging heavily on the White House Executive order. We are following a lot of comments.

We're part of the NIST AI Safety Institute, uh, which is a great initiative by the Department of Commerce. This is where more than 200 leaders entities are coming together to define best practices, tools, guidelines on ai, including on security. Uh, and if we dive in, you know, there, there is a couple of things that I would like to highlight.

So we have those foundational documents from NIST that were developed for software, perhaps SSDF, the Secure Software Development framework is the most prominent. Those are gonna get reopened and revised to address both how AI should be secured as code, but also how AI is impacting security as a vector of attack. So SSDF revision was just released.

We saw, um, I think it's, it's, uh, sorry, I'm sorry. It's the, it's the AI Risk Management framework, uh, appendix that was released for generative ai. And we are starting to see those documents being released and revised.

0. It's considered one of the most important compliance documents for cybersecurity period. And the way it's gonna work is that the cybersecurity framework is leaning on the AI framework, the AI risk management framework from NIST for the security elements of AI and kind of referring to it.

So there is a nexus between them. And again, all these AI documents are getting reopened right now as we speak. So a lot of the work we do is we, you know, scrape what's coming from nist.

We connect to companies interest area, and we allow them to actually work with NIST against feedback. So the NIST team is actually here at RSA, uh, and we have a lot of government officials coming in from RSA, and, you know, perhaps, uh, a message to y'all, you, you can be part of the conversation. You can file to comments to nist, you can engage.

Um, there are a few themes that we really see across regulatory requirements for ai, but the most important one, and it's kind of common between the UA Act and the US, is the importance of governance and compliance. So the way it works, it's not enough to secure the AI and have practices for teaming. You should also thinking about what are the artifacts?

How are you measuring this is working, but also what are the artifacts of compliance you're gonna be producing? So for example, in Europe, for high risk ai, we have this concept of conformity assessment. Conformity assessment, which is all about not just how you secure, but how you demonstrate your securing.

So this concept of governance is a key pillar across all the AI regulation. Another key concept is retaining and testing, uh, with, you know, kind of the hacker mindset. Um, you know, this is, this is common in many requirements.

It's a big focus of the executive order. It's a big focus for this work. Um, so your organization should be already obviously working with, uh, friendly hackers as my sister calls them, and collaborating with vulnerability disclosure programs and thinking about bug bounties.

But that concept is gonna get pushed farther even with ai. And what's striking is that the traditional concepts for ai were focused on the notion of hackers finding security vulnerabilities, and now they're finding bias issues. So we are actually working with governments to make sure that they have the right protection to do that, uh, testing.

But that's gonna be a, a common, uh, theater, um, data. Um, we, we, we actually say it's AI regulations, but the common theme between security privacy and AI regulations is a focus on data that the governance data in context, data security, a big opportunity for privacy, for privacy enhanced technologies. And above all, this concept of the data inventory and the AI inventory, that's the first thing you're gonna see a lot of push for.

Because as in supply chain, we know the foundations of understanding how to secure is understanding what you have, what you have, and what is coming for fourth parties, what is foundational with what the AI is interacting AI in context. A lot of that is inventory management. It's a big common theme, um, in the regulatory requirements.

Um, not gonna go to all the details here because I kind of gave you the high level. Um, I will say, you know, again, opportunities for companies that are doing, uh, solutions for ai, we work with many of them. We think AI governance is not gonna be an easy task to achieve.

So we are really empowering security companies to think about not just how they help, you know, their ecosystem to secure for the most important attacks, but demonstrate it to the regulator, demonstrate that they deploy those solutions and, and moving the needle. Um, Europe has a bunch of AI use cases that they're calling, they're basically saying are too high risk to be deployed. And in the high risk, um, kind of, um, bucket, they are folding in.

A lot of, uh, ai, AI is in medical devices, AI and vehicles, AI and recruitment. Recruitment in education. And the common theme here is what is the impact of the ai, the ai, what's the context?

And I think a lot of what I'm excited about in this RSA is mesh, like really allowing organizations to understand the problem, even from a compliance, by understanding the context of the ai. This AI is being used, but it's used in the context of my own data or high risk data because it's associated with decision making on medical or, or kind of employment implications that's gonna have all other layer of regulatory requirements. So we already know this need this, we already know this from privacy, this idea that we have PII or we have data that is in the context of hipaa.

So we have more requirements. That mindset is going to ai, which is all about understanding which AI is being used and what's the context. That's kind of a very important pillar of it.

Um, and you know, what I would say is you look at the security elements, you see this like transparency that a govern governance, human oversight, a lot of cybersecurity requirements. I was just talking with the European Commission. We're gonna have perhaps one system of compliance between the use Cyber Resilience Act.

So how many of you have heard about the use Cyber Resilience Act? I know one person here that worked with me at Intel, so she knows, uh, a few hands, uh, that is the most comprehensive product security regulation ever introduced in two, in three years. The product is not coming into U Europe, not coming into the customs.

If the, if the product cannot demonstrate self at the station, in some cases third party certification, a set of security baseline requirements, vulnerability disclosure program, s bomb capability to patch no default passwords, we are really experiencing this avalanche of requirements coming from the enterprise to the product. So the EU is thinking holistically on these issues between NAS two directive between the CRA and the UA Act and the security infrastructure there. They're thinking about one, uh, compliance mechanism, uh, big penalties.

So no surprise there, you know, 7% of of your annual turnover, just enough to kind of get your attention. And I know we've been kind of run running fast here. Uh, so I'm not gonna go into all the details.

I'll just say I'm really proud, um, to be working with those innovators because we are making real impact. This is from last month, a Department of Justice letter that went to the Copyright office as part of our advocacy for Safe Harbors for AI re teaming. We have protections for hackers under the DMCA from copyright law for security testing.

But we need those to be expanded to AI findings. And, you know, we are filing comments and engaging, it's a lot of work. But then Department of Justice says, oh, we agree with open policy.

This is a concern. We would like to see this type of an exemption. And it's important because we have hackers here in the room and we wanna make sure that we have the best find testing AI and deploying it and feeling like they don't, you should not be concerned about legal ramifications and concerns.

And with that, thank you my friend. Thank you for having me. Fast ride as usual.

Uh, but I'm, I'm around for, for questions. I appreciate you.