Sponsored: AI is attacking… it needs to defend, too

0:04James Wilson: Hey everyone, I'm James Wilson and welcome to this Risky Business sponsored interview with Prowler founder and CEO Toni de la Fuente. So AI, coding agents, LLMs, whatever you want to call them, they've pretty much put the ability to create an app into the hands of seemingly everyone and anyone by now. But where are all these killer apps? Where's the next generation of amazing apps that surely are being created by all these people that now for the first time ever can unleash all their ideas? Well, the answer to that is, at least partially, is that a lot of these killer apps that are being vibe coded, they exist only within a particular organization.

0:45James Wilson: because these apps, they solve a problem so specific to that organization that often the idea of turning it into a product or a SaaS just wouldn't make sense or the old juice wouldn't be worth the squeeze. But you know, that vibe-coded app that is now the centerpiece of productivity or a competitive advantage for an organization, it's still got to be deployed somewhere to be useful. And so while anyone can vibe-code an app these days, is that doesn't necessarily mean that an agent is going to know how best to securely deploy it. In this interview, Tony and I talk about how the way that we're securing infrastructure, it needs to evolve to match just how vastly different the art of creating software is these days.

1:32James Wilson: You know, it's amazing that even just leaving the IDE or even the agent UI is seemingly just too much friction. We just want to do everything within that agent app, within that harness. But ultimately, what we're talking about here is this is not about Prowler versus other more monolithic solutions. In fact, as Tony explains, in terms of deploying your applications and infrastructure in this modern AI era, it's actually about the right combinations of tools integrated right where they will have the greatest amount of leverage. I'm going to drop you here into the conversation where Tony picks up from my queue where I said, look, isn't it amazing that everyone can write software these days, but so little of that software seems to be getting out in the public and rather the next generation of killer apps are actually all internal.

2:20James Wilson: And so how do we think about securing those? Enjoy. It's a very good point because you're talking about building applications super fast now with AI and using internally. but internal applications are no longer internal, right? Are in the cloud, in any cloud. And when I talk about cloud, I'm not only talking about AWS, GCP, Azure. I'm talking about many other cloud providers, but also many other SaaS providers for infrastructure. Let's say Cloudflare, let's say Vercel, let's say Microsoft 365, because you are using authentication with EntraID, right? Or many others. So when you create applications for your own company, nowadays, people working from home, big, small companies, everybody is distributed.

3:11You need to put something somewhere. And that where is usually the cloud. And people is taking care about the functionality of that killer app, but not the security.

3:21James Wilson: Right. So why? Because not everybody can afford certain solutions. Right. And it's not only about being able or not to afford certain solutions. is being able to assess the security continuously or the continuous compliance of your just built application the same way that you built the application with your agentic way, let's say, right? With using agents. I think that's the interesting thing here is that it's not just that AI has created the ability for anyone to develop software. It's created a different approach to developing software, right? Think about that. We used to have the days of like huge ticket queues.

4:00James Wilson: And if you had ideas, you'd put them in the queue and someone would triage them and you'd come out with a working set for a sprint. I think a lot of that's gone by the by. And we're just we're very much in the moment of thinking, what's the need right now? What's the thing I want to add to this app? How can I have a conversational discussion with an AI agent that produces the feature that will solve the problem that's ahead of me right now? And I'm curious as to how that sort of different way in which software is being developed or being thought about also translates into how you think about the right fit for purpose security tooling right because something like a large monolithic off the shelf does everything security appliance that for a security solution for cloud security posture management and all things cloud safety that feels a little bit at odds in terms of the scale that it operates the pace that it operates the way it would integrate versus the very much in the moment i've got a problem hey hey, Claude, hey, Codex, solve this with me right now.

4:56James Wilson: What's your thoughts about how that determines product market fit? The new way of building applications from an AI-driven IDE, let's say Cloud, right? Cloud Code, for example, is so you create the application, the multiple modules, the multiple options to deploy, and then what's next? Of course, you need to maintain, update, upgrade, but also secure and make sure everything is in the loop. So our approach is different to others because we want that practitioner, the developer, that creates the application but in many cases it creates the infrastructure as well underneath. They can use the same flow, the same experience without leaving the IDE.

5:44So with Prowler you can of course find the cloud state, the identity, the configuration, the vulnerabilities, prioritize for attack paths, the blaze radius, etc. Tell the agent, hey, explain this to the user because we have things to do here. Then Prowler can plan the fix because we have a very large database of deterministic artifacts of cloud security for detections, remediations, compliance frameworks, etc. Then the agent can execute. Okay, Prowler says that I have to fix this, I'm going to fix it, because the agent has access to the infrastructure code, or even the application or the calls to those cloud providers, right?

6:27And then Prowler back on the game to verify and tell you, hey, you are secure this way and you are compliant, for example, you have to be SOC 2, type 2, etc. So that relationship between what Prowler can do in the cloud, that is the deterministic database, is the knowledge, is the prioritization, all the different signals that we get from the cloud to tell you this is important, along with your tools, the new developer tools, right? your agents building and doing the things right. So to close that circle is what we call the agent-friendly cloud security with Prowler.

7:09James Wilson: Right. Yeah, it is fascinating how any time a developer or a product builder, VibeCode, or whatever you want to call them, has to move alt tab or command tab out of their IDE or out of their agent. That's just so high friction, right? Everyone wants to be in that agent, You know, even sometimes with the app alongside and running in the sidebar of the agent, because it's just such a rapid development cycle. You know, not discounting what you said there in terms of Prowler can be in that session as well, helping you diagnose and fix things. To push back a little bit, everyone's got an MCP server these days.

7:46James Wilson: so why you know what what is it about prowler's approach here that perhaps sets it apart from from any other you know cloud auditing tool set that i might otherwise be able to get my hands on yeah the point is it's not only about the mcp which is like okay you have an api it's about what tools are behind that mcp right and what is the capabilities what is the speed or what with everything that you can do. So, and we offer from our MCP, once you connect your IDE to Prowler MCP, Prowler Cloud MCP is all the Prowler documentation. So your IDE also becomes an expert in creating detections or remediation if you have to create your own providers, your own detections, remediations, et cetera.

8:31So that is one of the things. So it becomes a cloud security developer. Second, it becomes a cloud architect because it knows everything that is in your cloud because Proder has all the findings and all the relationships between all the resources in your cloud. And also it becomes any other cloud expert because it has access to our documentation as well. It has access to all the information that also the model has inside. So we do have the MCP looking at everything that any engineer can do. but if you think about nowadays the new infrastructure the new way of deploying infrastructure is as I said at the beginning it's not about just okay deploy everything into AWS it's about deploying everything in the cloud whatever that means and nobody can know about everything nobody knows okay what are the best practices about deploying this in Vercel or what are the best practices of deploying this in GCP nobody can I mean you probably know one cloud

9:34James Wilson: to clouds but Prowler knows about all the clouds yeah so you make your agent an expert in many many clouds and also being able to create other and add new cloud providers into Prowler to grow the the set of of artifacts actually more than 20 percent of our artifacts are created by the community and that is something that that you cannot build with with it during a weekend i mean it's it's a community is something that is belongs to people right belongs to our users and belongs to our needs so you need this and you are able to create it and now with ai you can create so many artifacts for prowler and that is that is part of our main differentiation right yeah so we've talked about how, you know, Prowler can sit alongside you as you're building that killer app that's completely internal, custom bespoke and does amazing things for that organization.

10:32James Wilson: I get that. The flip side of this that you just touched on is there's also, I think, a pretty big change in the way both applications are being deployed, but also the risk and the attack surface, right? So let's, you know, there was for a long time, you could sort of take comfort in the fact that, you know, the small cloud app that I've deployed, if it's going to take an attacker, I don't know,$100 ,000 worth of effort to attack me, but they're only going to get $10 ,000 worth of valuable data out of it, then there's probably no real risk here. The other flip side to AI is it's made launching very large scale attacks very cheap, very fast, and, you know, largely and increasingly autonomous.

11:13James Wilson: And so I think that changes the equities here. And I'm curious as to what you're seeing in terms of, you know, where do these coding agents like to deploy these custom-built, vibe-coded, internal killer apps? Are they being deployed safely? Are they being deployed securely? Are people even thinking about that? And then, I guess, what's the new paradigm that's emerging for how these killer vibe-coded apps are being deployed by organizations? What I see every day is that people is worried about the problems they are solving with their apps, with their killer apps, but not everything else around those apps, right?

11:55And as you said, so now in a time where attacks are automated and driven by AI, you need to defend automated and driven by AI. And that is what we are trying to tell people, right? So remember the old days of infrastructure as code automation. I mean, when that started, when the cloud started with CloudFormation, Terraform, right? We were like, okay, we can deploy infrastructure super fast with infrastructure as code. So we need to secure infrastructure super fast now because we can no longer do hardening or protection or defense as before infrastructure as code. Because otherwise, you are late.

12:40So if AI is attacking, AI has to defend. And that is key on what we say and what we do, actually. That is why we wanted to be that broker, let's say, or middleware between everything that we can access to have a central location of authentication with the right permissions, the right provisioning and the right list of checkpoints. That is the deterministic base. And then your AI will take advantage of it. Because if you ask any AI, hey, go to GCP and do security, it has to figure out what to do security is, actually. Yeah. And that is a problem.

13:24James Wilson: Yeah. Okay. So is it safe to say AI is allowing anyone, I guess anyone in inverted commas, to write an app, to even deploy an app, but not necessarily secure an app? And security is not just at the time of build and deploy. It's like an ongoing concern. So I guess what's the recommendations you give to people about how do you keep an eye on things beyond that point of, hey, you know, the app's deployed, we can log in, let's go. yeah so i mean the basics are of course make sure you have a mfa or even with single sign-on all the stuff in your app in your app that also the access of your infrastructure is under control um not to use the admin and all the basics are not that basics so we see that happening every day every day so that is why deterministic security helps a lot and doing not only the basics sometimes are the basics but also and connecting dots and signals are able to to give you a you are going to have you may have a potential attack this this way or that way what what others i mean we can call like um exploitable signals which is an interesting point before we got talking we were saying how Even if you had all the tokens in the world, you're not going to go and create the next crowd strike necessarily, even with AI.

14:53James Wilson: For reasons of AI is really great at solving the immediate problems you have in front of you. It's not necessarily great in creating a solution for all 10 ,000 problems that millions of organizations around the world are all going to face. but also it's just, it can't generate, I don't think it can generate the history of rule sets and the incredible amount of knowledge that gets built up over time. And I just, I wanted to test, is that what you mean when you say deterministic security sort of meeting, I guess, the boundaries of indeterministic coding agencies and what's the benefit that you see there?

15:29Well, there is something very important here that we do and actually we do with our community is the open source foundation of what we do. And that allows the community of cloud security community in general, but also the Prowler community to grow the different capabilities that we offer, right? And the capabilities that anybody needs. So if a company, whatever, needs security in their cloud provider and is probably not in Prowler yet, they can create it very easily. So with AI, you can create any provider in Prowler in a matter of minutes, literally. And what is to create a provider and why that is important to be open source?

16:14Because first, detection is a commodity. Cloud security detection is a commodity. Everybody should detect everything. So what is not that community is, of course, the prioritization, the fixer plan, you know, the execution, the verify, the continuous monitoring, you know, the infrastructure to do all that stuff. But detection has to be, belongs to the community, let's say, right? Because it has to allow you, if you have a new service, which happens in many, many cases with major cloud providers, like, I mean, remember the days of Bedrock or whatever new service that comes with some security options, let's say that way.

16:55You need controls to make sure your deployment are secure or are following the best practices. It's not only for compliance reasons, but also to prevent security incidents. And being able to extend the capabilities of a solution super quick, based on your needs, and make those capabilities open source or just private for you, that is something needed in the community. So we are full of black boxes all over the place. And we see everyday users or customers using Prowler to validate what others are doing because they trust Prowler, because the community trusts Prowler. You can go to our GitHub and you can see all our contributions and all the new controls and everything is something that you can check.

17:46Also, governments also trust Prowler for that reason, instead of just waiting and expect that it's going to work.

17:54James Wilson: Yeah, well, I think that's a great place to wrap this up. We know, as we've said, AI is increasingly making it possible for anyone to develop those killer apps. Those apps aren't necessarily getting out there in the public domain. They're often very internal to organizations, but that doesn't prevent the need to be thinking about security all the way along the development process and then after it's deployed into production. Tony, it was great to spend some time with you. Thanks so much for dropping by for a chat. Thank you, James. Thanks for having me.

18:27you