AWS Financial Services Symposium 2025 | Paul Davis, JFrog

Clips
More from AWS Financial Services Symposium 2025

Paul Davis

Field CISO

JFrog

play_circle_outline JFrog's role in software supply chain and security

play_circle_outline Maximizing Software Security Through Visibility, Traceability, AI, and Machine Learning Technologies

play_circle_outline Enhanced Security with AWS Integration & Continuous Scanning for Zero-Day Threats

play_circle_outline Challenges of agentic AI and code generation oversight

play_circle_outline Discussion on agentic AI and MCP for workflow automation and security

play_circle_outline Impact of compliance regulations on AI development and liability responsibilities

Info
Transcript

Paul Davis, JFrog

Paul Davis

Field CISO JFrog

In this conversation at the AWS Financial Services Symposium, Paul Davis, field CISO at JFrog, joins theCUBE Research’s Dave Vellante and Scott Hebner to explore how financial institutions are scaling operations and modernizing infrastructure through cloud-native architectures and resilient DevSecOps strategies. Davis underscores the urgency of maintaining high performance and regulatory compliance amid growing transaction volumes, rising CVEs and evolving AI technologies.

Davis dives into real-world tactics for securing software supply chains using ... Read more

explore Keep Exploring

What are some key points of focus for integrating security into AWS tools and platforms? add

What are some of the emerging risks associated with the rapid acceleration of AI technology, particularly in terms of cybersecurity and malicious intent? add

What are some ways in which partnerships and continuous scanning can help provide additional security and reduce risk when utilizing open source models like Open Force, TensorFlow, and PyTorch in the financial sector? add

What are the potential risks associated with agentic AI and why is human oversight important in this context? add

What are some concerns and considerations around the development and use of agentic AI and MCP technology? add

What are some recent changes and challenges in regulatory compliance that businesses in the financial sector are facing? add

bolt Powered by CUBE AI

Paul Davis, JFrog

search

Dave Vellante

>> Hi, everyone. Welcome back to New York City. This is Dave Vellante with Scott Hebner, and you're watching theCUBE's coverage of AWS Financial Services Symposium, the 10th since 2015. This company, Amazon Web Services, has really ascended in the financial services community and we're really pleased to have Paul Davis here, he is the field CTO at JFrog. Thanks for making some time for us, Paul.

Paul Davis

>> It's a pleasure. Thank you. Glad to be here.

Dave Vellante

>> Okay, so we're talking software supply. Be honest with me, when was the first time you heard the term software supply chain?

Paul Davis

>> Today.

Dave Vellante

>> Really? Okay. So it goes back to the early part of this century, and Gary McGraw's work, I don't know who he is, Maven Central, you may know where that is, largely confined to security conferences and trade journals. And then 2014 to 2018, the target HVAC breach, that was 2014, and the Equifax Apache Struts incident, 2017, introduced the idea of supplier risk. And the 2019 Capital One cloud config breach, blah, blah, blah, NIST. But when I first heard about it was SolarWinds, which was 2020. So December, 2020, malware inserted during the vendor's own build process. Here you are thinking you're patching, doing the right thing and this insidious malware comes in and then President Biden's executive order and then the Log4J, and then it's become really a critical part of the cyber security conversation.

Paul Davis

>> Yeah, I mean, the thing is is people don't realize that 80% of the software that is used to build software is actually written by somebody else. We talk about third party code and first person code. That 80% means you're pushing 80% of your risk to somebody you don't really know. And so you've got to have controls around that. And obviously security people have a little problem where we can't say no to risk. If there's a problem and nobody else sticks their hands up and says, "I'll protect you," the CSOs typically go, "Yes, I'll do it." And so supply chain became that. And then we talked about vulnerabilities and CVEs and all that sort of stuff, that raised the profile. And the fact it's business impacting, it's almost like we were surprised that all of a sudden there's this new threat area that we weren't thinking about. And so it becomes complicated.

Dave Vellante

>> And the way historically a lot of organizations would deal with it was paper, like contracts that say, "You won't do that, right?" "Yes." "Okay, great," and just like a checkbox item and you'd report that up to the board. "Oh, yeah, our supply chain is secure," and reality, it's not. And cyber insurance came into the fore and they said, "No, no, no, this is not okay." We still haven't solved the problem, but that's what JFrog is all about.

Paul Davis

>> Yeah, I mean, I've been in the process where being a CISO for large organizations and a lot of them, they'll just sign the policy standard guidelines that they wanted to follow without actually reading them, without trying to understanding the obligations. And now in today's world, people are realizing they've got a risk.

Scott Hebner

>> So in the old days, long time ago when I was the product management leader for Rational Software, if you remember them.

Paul Davis

>> Oh, my. Wow.

Scott Hebner

>> I was at IBM, we used to talk about the software component life cycle.

Paul Davis

>> Yes.

Scott Hebner

>> So is the software supply chain different than that? It seems like this has more of a-

Dave Vellante

>> Definitely....

Scott Hebner

>> regulatory and security tracking, audibility versus managing the actual-

Dave Vellante

>> Well, it's like you're saying, you've got open source, you've got third party software, you've got components, you've got microservices, you got custom code. I mean, that's-

Paul Davis

>> I mean, all really does come down to the business impact. I focus on the business impact of bad software, whether it's trust, regulatory compliance, just the ability to stop doing so many changes, change windows or incidents or audit findings. Those are all business impacting. And it's only now like our aware... It's not only now, but recently in the past eight years, we've realized that it's become a problem and now we are dealing with it. And thankfully a lot of organizations are dealing with it and starting to look about how they can protect it. But we have had this terrible capability or challenge of doing point solutions. Oh, I've got this problem, buy that, problem this. Now we've got all these multiple tools, whatever, which is really chaotic because the more tools you have, the more complex it is to actually measure and track and manage risk.

Scott Hebner

>> Imagine it gets even more complex with the world of AI agents where they come from everywhere. And so the visibility and the traceability of how all these things come together, not just agent to agent, but all the underlying software that I could imagine, it's all over the place. Different origins, and you need to understand what you're dealing with.

Paul Davis

>> Yeah, I mean, a lot of times when we talk about security, we used to say, well, the problems happen when we involve humans. Now with AI, we've got this problem of we've got now another type of thing which can do automated actions that we're not aware of, and how do we control that? I was a CISO, I used to have a rule, which is three strikes, you're out. Three security things and you're out of wherever I am. You can't really do that with an AI agent. You can't fire the AI agent. You can shut it down, but then business says, "That's a problem, we need it back." And so how do you protect yourself against that?

Dave Vellante

>> So this idea of tools creep with our partner, ETR, we track this stuff. We've done now two annual surveys just ahead of RSA, and we found that there's least some evidence that tools creep is slowing down. I'll put it that way. There's a big theme in the industry as you know about consolidating tools. Only about 6% of the, I don't know, 1800, maybe it's not that big, maybe it was like 800 I guess folks that we surveyed said they were able to consolidate their toolstead, the number of vendors in their tools stack, and it's the security stack. But there is evidence that it's slowing down. And then you have these big companies, certainly Microsoft, Palo Alto, CrowdStrike to a certain extent, Zscaler maybe partnering with a CrowdStrike, trying to build that consolidation team. Brings me to AWS, which is interesting because unlike Microsoft who really monetizes security in a big way and competes very heavily with the likes of CrowdStrike and Palo, AWS really more focuses on the infrastructure. They have Security Lake, they have Security Hub, but they leave a lot of room for third parties to add value. So how do you think about the AWS ecosystem? Where do you guys fit? How do you fit into some of the infrastructure, I'll call it, that AWS provides?

Paul Davis

>> So just like you talked about those other vendors producing this consolidation, this path for consolidation, so JFrog is doing the same thing for developers. So we call it like the platform play, the end-to-end. Important thing, we need to have process because a couple weeks ago said, "Platforms won't work unless you have process." Interesting.

Scott Hebner

>> Interesting comment.

Paul Davis

>> But from the point of view of AWS, we've run and a lot of our customers use our product inside their environment to get that single point of truth around binaries. So we had to tell whether they're malicious or safe or whether there's licensing issues or operational risk. But it enables you to get that new perspective of end-to-end platform. As I call it, layers of defense, sounds familiar to the other products, but I can do layers of defense now for software development. So from the design phase through to the developers through to, as it starts to roll into production. And so providing that framework means that we can take the other tools and integrate them and remove those gaps in the visibility and get what I call the traceability. So I can actually see where a package comes from, who developed it, and it's what I call genealogy. It's life history of how it got into production, what was approved, what was blocked, is it safe? What were the other evidences that were associated with that particular release? Very important.

Scott Hebner

>> So when you take your SDLC, you plug in JFrog, perhaps you use AWS L-Post.

Paul Davis

>> Yeah.

Scott Hebner

>> And then not only within that one SDLC, but it can support a hybrid environment across your enterprise where things are developed in different places. Is that the idea of how that all comes together?

Paul Davis

>> Yeah. There is this need to have sort of a single point of truth regarding your binaries. It just reduces costs, it improves speed and also reduces risk. So we support that. So when you say points, we support over 40 different packages now package types. So the fun part is the transformation of getting the teams to migrate to that. And a lot of customers are doing that now, but it's still an ongoing thing that has to happen. And you talk about the tools, consolidation. I spent a lot of time working with customers, helping them understand where their tools are, where their mapping is and where the gaps are.

Dave Vellante

>> So don't hate me for asking this question, I apologize if it's like trivial. Would you put yourself in a posture management category, or is it more sort of shift left security?

Paul Davis

>> Okay, because we've got a broad journey, it's shift left and shift, right? But I wouldn't say an ASPM posture management. However, the data that we have-

Dave Vellante

>> Can support that....

Paul Davis

>> can support that. First of all, I can tell you whether this software is supposed to be there. Is that a rogue app or is somebody going to tweet it or bypass the process? Because you know, in security, we trust... Or we used to say trust and verify. I don't follow that. Verify. Verify and maybe trust because I'm paranoid. Okay?

Dave Vellante

>> Don't trust.

Paul Davis

>> Okay, don't trust. But you need to have that ability to have that. So an audit comes on and says, "Where did this application come from?" You can go back and say, "We've seen, here's the build process. Here's all the stages. Here's the quality control scanning did, this is how it rolled up into the infrastructure. This is the data feeding into the AWS tools, et cetera." That all feeds there, so you can have a more 360 sort of model for... Because remember, security for me, ignorance is not bliss. It's the unknowns that hurt.

Scott Hebner

>> So how does that relate to the AWS security?

Paul Davis

>> Well, we can feed data into that, right?

Scott Hebner

>> Got you.

Paul Davis

>> Because the thing is, it's like if you want to make a CISO happy, say, "We can give you more visibility." And the one thing, I mean I've met over, I don't know, a large number of CISOs last year, and we all want to work with the developers. We all want to get the data there. Because we're talking CVEs, we're talking about risk. And a lot of them are sort of trying to get integrated with that development life cycle and get the developers to become effectively security and extension of security.

Dave Vellante

>> So I want to push you on that, because I feel like when cloud really ascended, it became the first line of defense. And then people realized, "Oh, this is a shared responsibility model. Oh, I can't just put my data in the cloud and sleep at night." And so I think it was not well understood by many organizations early on. And then to your point, the CISO was essentially responsible through the development organization for securing the data and the apps. And so the app dev folks had to become sec ops pros, which they probably didn't want to do, which is a great business case for JFrog, right? And then you got audit looking over your shoulder. It's kind of the last line of defense. So the question is, what's the industry doing generally, and JFrog specifically, I think you've touched a lot on this, to make the CISO's life easier and the app dev life easier?

Paul Davis

>> I go back to data. It's funny, I'm seeing an interesting trend now where a lot of times the dev sec ops model has been run by the development team. It's kind of fun. But now security teams are starting to say, "Hold on a second. You're defining policy? You're doing waivers? Excuse me. No, that's what we do." And so they're starting to get sort of almost one might say territorial about it and saying, "No, no, we have to work together on this. I have a risk register. I have to report risk to the board, et cetera. We need to work together. You can't do this independently." And that kind of synergy between the two organizations is happening. It's funny, I was at another conference recently and I was a bit cruel. I took a developer and convinced him he was a security professional in 15 minutes. He hated me. But as I describe it, developers are good hackers. You give them a technology, you give them a problem and say, "Find the solution."

Scott Hebner

>> So as part of that software supply chain, the machine learning models, the LLMs, talk to me a little bit about the AI part of this. .

Paul Davis

>> Well, yeah, the acceleration of AI has been fun to grip your hands onto the edge of the thing, keep your hands and grip there. The thing for me is is that the attackers have realized that the developers and now the data scientists are people they can attack. So now we're starting to see where the explosion of new LLMs every day, that some of them are turning bad, malicious. Whereas if I get my marketing, I'm weaponized LLMs. But basically just the act of downloading an LLM can cause it to launch malicious code. Just the fact that they're doing an install can cause that. Then you add on to it the fact that these models were all over the place and a lot of companies are still trying to work out where it's used. And so that whole thing of traceability, knowing who's doing the work and not being surprised is a key thing. And so first of all, we have a threat intel team that last year started doing the scanning of Hugging Face LLMs, right?

Dave Vellante

>> Oh, boy.

Paul Davis

>> And we discovered a number of bad ones, and we've also proven that we're good at it. We publish it. In fact, recently we just announced a partnership with Hugging Face, so we're now one of the sources of Intel. So when you go to Hugging Face saying this is malicious and then we show you all the date behind it.

Dave Vellante

>> Because ignorance is not bliss.

Paul Davis

>> You got it. Hey, I've converted you. Are you a hacker or not?

Scott Hebner

>> That's great because what they got approaching a million models sitting out there and open source and stuff like that. Well, so how does this all work with Amazon SageMaker?

Paul Davis

>> We are fully integrated with them, to be honest. And so we have that partnership to help facilitate that. And it's a growing partnership where we're adding, I can't talk about some of the things coming out, but that partnership is growing, so we can provide additional security

Scott Hebner

>> Which allows you to tap into the open source models on how interface with a greater, or I guess a lower risk profile.

Paul Davis

>> Yes. The other part I must add in is that if you think about the libraries that support LLMs, there's Open Force, TensorFlow, PyTorch, and we have seen situations where they have turned bad or they've got vulnerabilities or zero days that people don't worry about. And that's one of the things we can do is there is the requirement in the financial sector to do continuous scanning. It doesn't mean we literally scale all the time, but we update our database multiple times a day and we can say, "There's a new zero day come out in your package that you've just pushed six months ago, a year ago, two years ago. How do you respond?"

Dave Vellante

>> Okay, so it's not 24 by 7 by 365, but you are taking granular snapshots.

Paul Davis

>> Yeah, multiple times a day, we're doing updates to that database.

Scott Hebner

>> So our conversation here at this conference about agentic AI, you're empowering business users. There's low code, no code environments. So the notion of shadow AI or shadow agents or whatever, which obviously becomes part of your software supply chain. That's why I was going before on this whole notion of with the outposts and stuff, you can support a distributed hybrid environment. That seems to be an emerging problem, if more and more end users are creating their own software. Or workflow.

Paul Davis

>> Yeah. This is the fun thing. It's like I have a joke. How do you detect if source code is being generated by Gen AI? It's documented.

Scott Hebner

>> Yeah. With periods.

Paul Davis

>> So from that perspective, I think it's a challenge. I think part of it is, I've done a lot of work too using gen AI to generate code and it's getting better. It's like, recently asked it to generate code and it didn't include the password inside the code, it actually referenced an environmental parameter, which was like, "Wow, it's amazing." But I think this is something that is going to require discipline. The whole thing with agentic AI was two parts, code generation. You need to have human oversight. You can't just rely upon on pushing code out. Yes, it can accelerate. And as we've been hearing at the conference, we have seen developer efficiency grow, the ability to generate code. I find it invaluable. I can program in 12 languages. I can't remember how to write something and C and then move to Perl and then move to Python. I can just say, "Hey, convert this." And I read it and go, "Aha, add this in," and then it's good, it needs oversight. And I think that's valuable as an inspirational model. Then we've got agentic, which scares the living daylights out of me to be honest, because we need additional controls to monitor at a transaction of what are the outputs. And I have suggested recently that we need an AI engine to monitor agentic AI. So we need AI to monitor AI to see exceptions, because they're running at machine speed. Because in the financial sector, we're doing transactions. If we're doing a stock exchange, that massive transactions per minute. Human can't keep up with that, so we need something to have guardrails. But because a agentic AI has, every time it does it, it does it slightly differently, the outputs matter.

Scott Hebner

>> Do you engage with chief AI officers? Let me preface that with some thoughts that I've been hearing as I've been talking to people.

Paul Davis

>> Yeah, please do.

Scott Hebner

>> Is that the chief AI officer traditionally has been a business technical guy. More people are talking about, well, they're actually kind of the head of the digital HR department, for the digital work, which is a key way of saying that they're responsible for governing what engines get deployed. And that's not in the development or CI, right? It's in that organization, sort of a digital HR group. I would think this would be really valuable to make sure that any agents that get deployed across the organization, that end users can then put into workflows, meets all this criteria of traceability and visibility, right?

Paul Davis

>> I'm probably going to make a career, blocking method. I would hate to be that person.

Scott Hebner

>> Yes.

Paul Davis

>> It's like the average tenure of a CISO is about 18 months. It's been like that for years. To put my destiny in the hands of an agent and say there's not going to be a problem. And that's one of the funny things about security is many industries have a sort of contingency or risk budget security. Security, there is none. You get it right, and if you fail, out. So from that perspective, it's scary. So we have teams, we have a team focused on AI. We actually acquired last year, Quack. And we've been advancing that because we see a lot of synergies between ML sec ops or ML ops and dev ops or dev sec ops and that synergy. So we have a team that's got the experience and knowledge and we're growing that expertise. But the idea of, I love the . I think that is a brilliant analogy. It kind of scares me because I'm paranoid, but I think it is. But the fact that you can't always consistently predict what it's going to do-

Scott Hebner

>> Right.

Paul Davis

>> It's like, I mean-

Scott Hebner

>> Or how it's used by people that are assembling the workflows. And by the way, your job is to be paranoid, so good.

Paul Davis

>> Yeah, .

Dave Vellante

>> I always felt like it used to be failure equals fire, and then that changed kind of mid last decade, people are like, "Well, they're going to get in." And so there seemed to be a little bit more patience. Are you saying that's changing now? That CISO's lifespan is maybe not shortening, but there's less patience?

Paul Davis

>> I don't know. To be honest, as a person who's lasted than the 18 months in here, and I have had some really fun incidents and where the actual failure became a success, where we were more engaged with the business, more involved in risk management. I think it's really how you present yourself. It's like one of the things I always say about security is, we're advisors. It's the business that has to accept the risk. Our job is to articulate as much the risk and say, "Look, this is what we can do," but in the end, we are not buzzing about this. "But you really want to do this?" And then at the end when they say yes, we have to let go and just put the mitigating controls in place.

Dave Vellante

>> And bad human behavior trumps good security every time, right? I mean, we've seen that-

Paul Davis

>> be human.

Dave Vellante

>> Right. Exactly. So speaking of agentic, I'm looking at a slide that Scott Mullins put up this morning. Financial institutions are generating value with agents. Some good logos on there, ADP, Moody's, Rocket, Robinhood, et cetera, Crypto.com. But they're only twelves. Now we know AWS. If this were Moterra, there'd be like 1200.

Paul Davis

>> Yes.

Dave Vellante

>> You wouldn't be able to read the slide. So it's early days still.

Paul Davis

>> Yes, .

Dave Vellante

>> But where are we in that?

Paul Davis

>> So we're kind of going through this sort of model of, okay, we've got gen AI, still learning that. Now we've got agents. Everybody loves agents. The whole industry, agents everywhere. Now somebody's come up with agentic AI and then MCP, which also scares me as well. Just means I can get on a plane, doesn't it really? But agentic AI is great from the perspective as I think if you have the guardrails in place and that ability to monitor, not just at the technical level, how many requests, what little questions, but also the outputs I think is key, I think that will make a difference. But I think it's still got a journey. Like the 808 protocol that was recently released, I love that because it's actually a secure protocol for agents to talk to each other. Now, the scary part of that protocol was also the fact it also supports long-running jobs. What is a long-running job? AI-

Dave Vellante

>> flow.

Paul Davis

>> Yeah. So that. And then you've got MCP from Anthropic, which also scares me because, have you heard of a product called Zapier?

Dave Vellante

>> Yeah.

Paul Davis

>> Yeah. It's like Zapier for AI. Yeah, it's plug and play, but there's no assurance around, that person who wrote that engine, did they do something good or bad? And as we mentioned this morning, perhaps you could yourself better than you could, than that person.

Dave Vellante

>> So I mean, MCP seems to be gaining at least some traction in terms of, I mean, MCP, there's a competing standard I guess that Google has put out. But at the same time, for agentic to work, you need things like MCP, right? So this is like-

Paul Davis

>> Well, I mean, yes, or you could write it yourself. Trouble it, it's a question of whether you trust it or not. How much risk-

Dave Vellante

>> Just mentioning you don't need it, it just facilitates greater adoption. And of course greater risk.

Paul Davis

>> Yes. You can move faster, but it's a bit like moving faster and then removing the brakes.

Scott Hebner

>> What it comes down to is just the reality that if you're going to be employing agents, and you look at different studies, McKenzie, Boston Consulting Group, just the name of a couple. Half the people that are deploying agentic workflows are going to acquire their agents externally from providers and wire them together. MCP is the way to wire them together.

Paul Davis

>> Yes.

Scott Hebner

>> You may be identifying a big area that needs to focus, which is ensuring securing, not only the supply chain of software that goes between those, but the actual MCP. Is that what you're-

Paul Davis

>> Yeah, yeah. That's what I worry about because I like to move fast. I like to develop fast. I've been doing machine learning for many years and I've been doing the agentic stuff and trying to get it to what it wanted to do and it doesn't yet. So I was still fighting that battle, to be honest. But MCP, yeah, I think it's a great step in the journey. But until got security, I wouldn't want... I'm shooting myself in the foot, but wouldn't want to roll it into production yet.

Dave Vellante

>> But don't you help address that problem with visibility and traceability? Is that part of your value proposition? This is this kind of chaos would be music to your ears, no?

Paul Davis

>> Well, we can sort of get the traceability, but what happens in production, we don't have those controls. We have to rely, and that it's funny, with AI and everything, there's so many new attack vectors. Filtration attacks, guardrails, , I mean just manipulation. I mean, it's a whole new set of tools out there. I was at RSA too, a couple of weeks ago and there's more AI tools than I wanted to think about.

Scott Hebner

>> Yeah, I know. So the fact that they're non-deterministic means dynamic, right? Which means it's kind of hard to monitor what's going on every day versus how you design it on paper.

Paul Davis

>> Yeah.

Dave Vellante

>> I was at RSA too, and somebody said, "Well, what about quantum? Are you concerned about quantum?" And the respondent said, "Hackers, they don't need quantum. They're doing just fine without it."

Paul Davis

>> I had a similar question, how does quantum... And I couldn't work out how it tracked into the traceability, but I do have some customers where they want to sign their binary so it only runs in their network. And if you could get that key and then insert your own binary, you'd be blind. Unless you had the controls to say, "Hey, this is a rogue app," because the checks haven't changed. But there's the query of quantum. I think the post-quantum cryptography is going to get interesting. It's a lot of infrastructure pain. I was sitting there trying to work out where somebody changed the key, it would immediately stand out in our model because JFrog's keyed on... Everybody has a unique set of keys, and so if somebody changes it, it immediately sticks out and says, "This is different."

Scott Hebner

>> Well, it certainly sounds like a big step forward. I just thinking about the integration with SageMaker, that you go out into Hugging Face and you can start selecting these open models and bring them in, and there's a much greater degree of security and traceability around that. That's a big step forward, I think, to addressing this, right? But MCP, that's another whole-

Paul Davis

>> Yeah, as I said, it's another fun subject.

Scott Hebner

>> It is fun.

Paul Davis

>> I must admit, one thing I don't do is I don't remember my dreams. So when I wake up in the morning, I don't remember the nightmares. So maybe I'm dreaming at MCP and I don't know about it.

Scott Hebner

>> There is a security element to it, but it may not be... You're sparking me to go look into it more. But the question is, I think you're more on how robust is it, one. And then the whole supply chain that feeds it of software or agents is another dimension of complexity here.

Paul Davis

>> I think MCP is a great thing. It's just we need controls.

Dave Vellante

>> Well, I mean, historically in our industry, it may not be the case for financial services, although it may be is, but historically, value creation trumps good security. People will go after the revenue and the productivity and the competitiveness and, "Oh yeah, we got to secure this thing." Maybe that'll change with AI, but based on this conversation, I'm not sure.

Scott Hebner

>> We're here at this conference today, services a lot of compliance and policy. So how does this feed into that?

Dave Vellante

>> Wow.

Scott Hebner

>> In other words, how is it contributing to these firms meeting their compliance mandates? Around AI in particular.

Paul Davis

>> I think for me, the financial sector has always been greater regulatory compliance. However, the regulatory compliance landscape is changing. It's like in Europe, they've got teeth on their things, the AI regulations. Did you hear about the product warranty update recently?

Dave Vellante

>> No.

Paul Davis

>> They removed liability and included software in product liability.

Scott Hebner

>> Oh, boy.

Paul Davis

>> And end of life, there's a whole journey onto that.

Scott Hebner

>> Fun time to be a software company.

Paul Davis

>> Yeah. So if you are doing business over in Europe, you've now got to worry about that. And I think the other thing we have a challenge is AI, it's a moving target because every state is coming up with its own laws around. It's like PII, but we have AII regulations and what we do there.

Scott Hebner

>> There was down in Delaware in the United States a couple of years ago, a fiduciary court ruled that corporate officers can no longer hide behind, "Well, AI told me to do it." They're personally responsible.

Paul Davis

>> Wow.

Scott Hebner

>> Which is going to drive even greater focus on compliance and policy and explainability. And so you're right. I mean, the regulations is changing all the time and keeps you guys in business.

Dave Vellante

>> Two-edged sword, right? Yeah. I mean, the companies that can afford to do it will do it. And some of the smaller guys that don't have the resources, they'll either try to skirt or they'll be in trouble.

Paul Davis

>> I thought you were giving me a glimmer of hope around the liability insurance that most CISOs have to have. That's not quite there yet.

Dave Vellante

>> Listen, Paul, thanks so much for coming on theCUBE.

Paul Davis

>> No worries.

Dave Vellante

>> Really appreciate your time and your insights. All right, for Scott Hebner, this is Dave Vellante. You're theCUBE's coverage of AWS Financial Services Symposium from New York City 2025. We'll be right back, right after this short break.