Securing AI Agents at Machine Speed With C1

[00:00:00] Your agentic AI might not be secure even with real-time data and proper guardrails. But Denodo makes sure your business has every avenue covered. By placing all your data platforms under one AI data layer, your business can reach semantic consistency safely and securely.

[00:00:21] So get your agents on the same page by visiting denodo.com and you can learn more about how to start trusting your agents to make business decisions. What happens when an AI agent completes thousands of actions before your next access review finds the calendar invitation?

[00:00:44] Well, in today's episode of Tech Talks Daily, I'm going to be joined by Alex Bovey, CEO and co-founder of C1. And together, we will discuss identity as the control plane for the agentic enterprise. He will explain why agents can reward Max beyond their assigned mission and why security controls must evaluate each action at machine speed.

[00:01:11] So we will examine how shadow AI credentials on employee devices, post-quantum vaults, runtime governance and human ownership behind autonomous systems are so important. And if you're looking for ROI on technology, Alex is also going to share how C1 increased weekly software mergers from 150 to over 1,500 with modest team growth.

[00:01:38] And you'll also learn how companies can replace blanket AI bands with a governed path that employees and agents can use without turning every experiment into a security incident. A lot of gold hidden in this interview today. And I look forward to introducing you to Alex right now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? Yeah. So my name is Alex Bovey.

[00:02:08] I'm the CEO and co-founder of C1 AI. I've been in identity and security for probably the better part of 20 years. And focused on identity specifically for the last 10 to 15 years. A little bit more about our company at C1. We are the access control plane for the agentic enterprise. So we help companies make that transition to becoming AI native through AI adoption, through automating access and securing identity throughout their company.

[00:02:38] Help them adopt AI agents. We like to say internally scale AI fearlessly, meaning we really give the tools and the technologies to our customers that help them deploy and use AI in their organization without being afraid by having guardrails and security controls in place that operate and move at runtime and machine speed as opposed to post facto. So that's what we do.

[00:03:04] We have great customers, ramp wrecks, Zscaler, folks like that, that they use our product on a daily basis to, uh, to help keep their companies secure. Awesome. And before you join me on the podcast, I was doing a little research on you guys and just reading about you, where you've come from and what you've created and the problems you're solving. Is that an origin story here as well? I always like going back a little bit and finding more about what my guests or what put the company on this path. Anything you can share around that origin story?

[00:03:34] Yeah, absolutely. So the origin story was my co-founder and I were both at Okta and we were leading zero trust products there. And I think our recognition at the time was we were building a lot of products and capabilities for sort of the human era, the zero trust era.

[00:03:53] And what we, we realized is that first of all, there were so many gaps and problems with that approach, even in the human era that we really needed a new identity stack that was going to operate at machine speed.

[00:04:07] And so I would say that was the original sort of thesis was how do we, um, how do we make it so it's dead simple to get these tools up and running so that customers get value, uh, so that they can be more productive by leveraging, you know, C1 to integrate with all their technology and give them visibility and help, you know, users and employees get all the right access.

[00:04:30] And then AI came along and frankly, just accelerated that vision and mission, uh, both internally with how we build products and go to market. Uh, but also in terms of the problem statements that we solve for customers, because in the AI era, you really do the, the ability to operate at machine speed and operate at the runtime really is paramount because AI agents are non-deterministic. They don't have a moral compass.

[00:04:57] They, what I like to say, they reward max, meaning they really pursue the reward and the task that's given to them. And they will go completely off the rails and do wild things to be able to accomplish those tasks and operating at kind of human speed doesn't work for that because human speed is at, you know, every 90 days, we're going to review the access that something has, or we're going to, you know, kind of, it's a lot of post facto evaluation. And agents can run thousands of actions in a minute.

[00:05:28] So how do you, how do you secure something where the ability to execute is so fast and there's non-determinism with what that, that machine or that agent is doing and they don't really have a moral compass so they can go off the rails pretty quickly. Uh, and that just necessitates and mandates the ability to really evaluate things more at the runtime in terms of enforcing security.

[00:05:53] And there's been a lot of noise that surrounds all things AI for the last three or four years. And this year in particular, it's all about agentic AI. We're seeing this rise of the agentic enterprise. I'm curious from everything that you're seeing to set the scene for where we're at now. Tell me more about this rise of the agentic enterprise that we're seeing. Yeah. I mean, I think we've gone through and well, we've done, we've gone through the agentic transformation ourselves at C1.

[00:06:21] We've built a cloud-based software factory. We have a platform for hosting internal vibe-coded applications. We encourage our team to use AI tools. We use our own products internally to enable our team to adopt AI MCPs and tools and automate, you know, manual processes. So I think going through that, that agentic transformation is essential.

[00:06:47] I would argue for every single business out there, really regardless of industry, because it ends up being a competitive disadvantage if you're not doing it. You know, I look at, for example, how fast our engineering team moves with our cloud-based software factory.

[00:07:03] We've gone from, you know, maybe six months ago, nine months ago, merging 150 pull requests a week, a pull request kind of being a, you know, a patch of software that you apply to your code base, to something like 1,500 merges. And that's with just an increase in our engineering team of maybe 10 or 15%. So that ability to execute and pull ahead, and that's just how we've done that on the software side.

[00:07:30] That doesn't apply to how we've done it on the go-to-market side, sales ops, marketing. Really agentifying your business just allows you to, I would argue, run 10 to 50 to 100x faster, depending on the use cases and the scenarios.

[00:07:46] And so it becomes a, you know, kind of an asymmetry in the market competitively, where if your competitors are adopting AI, and they're doing it successfully, and they've agentified a lot of their internal business processes, they are going to move 1,000 times, 100 times, whatever it is, faster than you are. And you literally just can't keep up.

[00:08:08] And so I think competitively, adopting AI and rolling it out to your teams is an absolute mandate for every single business out there, which is part of the reason it's such a board conversation. Every single CIO and CISO that I talk to is saying that it is literally the top board topic every single board meeting. And boards recognize the imperative here. And so they're pushing on senior leadership to say, what are you doing? What are you doing? How are you rolling out AI?

[00:08:35] How are you getting the tools internally enabled? How are you making sure your team is AI native? And of course, there will be people listening when they're hearing about agentic AI and individuals with hundreds of agents, teams with maybe thousands of agents. It does bring up a few security question marks there. And as a result, predictably, identity is now rapidly becoming the control plane for AI, not just another data source.

[00:09:03] But tell me more about that, the importance of identity and everything that we're talking about here as we move into this next phase of agentic AI. Yeah. I mean, I think fundamentally, there's a lot of different kind of arguments for where the control plane should sit. But I think, you know, network vendors will tell you AI is a network problem. Data vendors will tell you AI is a data problem. You know, so we're a little bit biased in this regard.

[00:09:29] But I actually do think that agents in particular is fundamentally an identity problem. And the reason why is it's all about access control. It's about governance. It's about enforcing the permissions and the actions that agents can take. And there's no better way to do that than actually at the identity layer, because identity is responsible for issuing tokens.

[00:09:53] It is the context with which you can actually govern the data that an agent can access and what they can do with it. And if you can't tie that back to an identity, either by giving the agent a standalone workload identity or by having a delegated authorization model where an agent is acting on behalf of a user,

[00:10:11] then you really just don't have a good way of controlling that agent and making sure that the agent has a well-scoped mission and that it's sticking to that mission and that it's kind of staying on the guardrails, if you will. So I think I actually think there's no better sort of surface area with which to control agents than identity for that reason. And one of the things that put you on my radar, a time where so many people are talking about how to secure agentic AI is you've created this.

[00:10:41] Well, you've launched the creation of the agentic control plane, which is incredibly cool. And you've brought four connected capabilities together to secure the agentic enterprise end to end, which will be music to many people's listening here. So tell me more about that, what you've released and what those four capabilities are. Yeah, so it does follow a little bit of the traditional sort of like protect, detect, respond type capabilities. But I would argue it's a little bit different. And I'll explain why as I go through it.

[00:11:11] But the first set of capabilities is around shadow AI detection and just getting an understanding of what's happening in your environment. And we call it shadow AI detection. But the reality is it's much broader than that. So we will look at and understand across cloud, SaaS, local agents on your endpoint, infrastructure hosted agents. We understand and we get visibility across all of those.

[00:11:39] We give visibility on local vibe coded MCPs. And we get visibility on credentials that are floating throughout your system. So the reason that's important, just to go a little bit deeper on that, is we've spent the last 15, 20 years as an industry trying to get software off of the endpoint. And with, you know, the proliferation of AI, you know, back to that comment about the board, boards are putting pressure on executives. Executives are pushing their teams to adopt AI.

[00:12:09] What's happened as a result of that is that every company out there has employees that are downloading vibe coded MCPs with supply chain issues in them with local credentials to enable the agents and the harnesses that they're running locally. That's a little bit of like a disaster scenario.

[00:12:54] There's a lot of them. And there's a lot of them. And there's a lot of them. And it's got built in governance. It's got built in workflows and automation and the ability to share it. It's got different levels of trust in it from a C1 perspective. So you can set up your vault so that it's kind of fully C1 trusted, meaning we can reveal secrets to more of a trust, no one architecture. We can just see metadata about the vault and we can't do anything else with it. So there's kind of a range of security profiles of which you can deploy the vaults.

[00:13:21] The reason that that's a beautiful technology is now we're not just stopping at shadowed AI detection and the ability to detect credentials. We can actually do something about it. And the thing that we do about it is we can actually vault the credentials and then we can inject them into the agent harness. We have an egress proxy. So we have a bunch of capabilities there to actually enable agentic workflows through that vault. So that's the second set of capabilities. The third is runtime governance for agents.

[00:13:50] And so when you think about if you look at this open AI hugging face attack, it's all about an agent sort of reward maxing, going off the rails, trying to do lots of different things to accomplish the mission that was given to it, including and not limited to hacking hugging faces infrastructure to basically understand like the answers to the test, so to speak, so that it could win its mission and task.

[00:14:16] And the way that you have to be able to prevent that is there's lots of kind of surrounding technologies. So one of the key technologies is runtime governance of what the agent is doing. So that's being able to understand the intent and the mission of the agent. What did we scope this agent to do? What is its task? And then being able to evaluate that in real time and at the action layer. So say, you know, this agent is trying to do these actions. Does that seem appropriate relative to the mission that we've scoped it?

[00:14:44] Or is it in the case of open AI executing a zero day vulnerability across an internal tools that it can create a message board that we're going to use to communicate with other agents? Like that is clearly not in the mission of that agent. And if you had the ability to kind of evaluate the runtime of what the agent is doing, that would be easily observable. And you'd be able to prevent or block that or bring a human in the loop to actually make sure that that's okay.

[00:15:12] So there's all sorts of guardrail capabilities there. Looking at the tools that the agent does, looking at the mission and the intent of the agent, being able to do kind of runtime guardrails around things like lethal trifecta to prevent, you know, data exfiltration. Runtime is pretty broad, but it's really about a set of capabilities that you can use to understand what the agent's trying to do if it's appropriate. And then authorize an agent on a per action basis.

[00:15:40] You have to operate at the level of the action in real time. And then the fourth capability is agentic security and intelligence. So this is really about we've gone through your environment. We've collected all this intelligence. We understand what your agents are doing and the behavior. We understand what potential security issues and risks exist within your environment. And we can surface that up in a digestible way that's explorable using our identity graph or through our findings.

[00:16:09] You can do automated remediation and detection of security issues in real time. It's really about kind of the threat detection response piece there. And that full story really is a broad set of capabilities that customers need to deploy AI successfully, to understand what's happening in their environment, and then to give customers guardrails or give their employees rather guardrails for how they use AI personally and how they deploy agents. I love what you're doing here.

[00:16:37] One of the things that stands out, I think, is if we look across the enterprise landscape right now, the enterprise now runs on humans and agents together. And that is only going to increase. I think that's one thing we can predict fairly safely. And, of course, securing all of that takes one control plane for every identity. And this is what you've created here. But I'm curious, from the conversations that you're having with customers and clients and different enterprises, is that message getting through?

[00:17:06] Are they still experimenting with the agents and humans together? Or are they actively seeking a solution here to try and secure everything and not just tack it on at the end? It's definitely – we are being actively sought out for it. And the message resonates with the market very well. Because, in part, we're also starting to see the evolution of compliance frameworks, for example, that are driving companies in this direction. So, the EU just put out an AI framework, for example.

[00:17:36] I don't think you have to be compliant yet. I don't exactly remember the timeframe. But one of the key requirements is that every agent needs a kind of responsible human behind it. So, even in these compliance frameworks, I think there's a recognition that you need a responsible party. You need almost like a human behind the autonomy that's responsible for the autonomy and what it's doing.

[00:17:58] And so, that message of one unified control plane that understands all of your humans and agents, understands the ownership between them, can delegate responsibility and authorization and do runtime enforcement, is really landing with customers. And I think even the early signs on the regulatory framework is that they're recognizing that this is an identity problem as well. And they're making sure that identity is kind of the main building block and primitive with which those agents are going to be governed.

[00:18:28] You made an important point now. I think it's important to stress. We're not talking about replacing engineers here. Their roles are shifting and defining problems, et cetera, in a completely different way. But their work is more important than ever. It's not about replacing them, right? I couldn't agree more. I think the Doomer narrative around AI taking everyone's jobs has been so overblown. And it's actually frustrating because we never saw it at C1, and we especially are not seeing it now.

[00:18:55] All that we're seeing is the acceleration of productivity from all the different folks in the business around how much they're able to get done by agentifying their workflows and using agents. So the kind of mental model that I like to give to the team is it's the nature of your work is changing.

[00:19:15] So it's not that it used to be in the SaaS era that your job was to be more productive by using more SaaS tools or by using more technology and software to be a little bit more productive. And I think what's shifted here is, but you were fundamentally doing the work. So you were producing the artifacts and the intelligence. And that was really what you were paid to do was to produce the intelligence or the outputs manually through your own work.

[00:19:45] What we're finding with agents is it's just shifting the nature of work where your job is not necessarily to produce the artifacts yourself. It's to build the factory that takes inputs and produces outputs and to curate the inputs and the outputs of those. So it's like where most people used to spend most of their time kind of in the middle, if you will. Now the work is shifted to the beginning and the end. You need to make sure you're prompting your agents to do the right work,

[00:20:15] that you're orchestrating your agents to have the right inputs. And then you need to be verifying and reviewing the outputs to make sure that they're good. And then doing loops if you need to on that work to make sure that it's kind of getting corrected if necessary. And so by not having to be limited by the number of keystrokes you do sort of in the messy middle, your productivity can just shoot up because you can do that exercise of managing inputs and outputs, I don't know, 100 times in a day.

[00:20:42] And you might 100x the amount of work that's being produced. It still requires human sort of like evaluation and review and making sure in curation, if you will, to make sure that what's getting produced is good. But just the nature of that work is different now. You're really not limited based on keystrokes anymore. You're limited on your ability to review inputs and outputs. Love that.

[00:21:07] And another area I know you're passionate about is helping enterprises adopt AI fearlessly with the right controls in place. And again, with identity as the control plane, just to hammer home what we're talking about here, what does this bring to business leaders listening here that are starting to think about this stuff seriously? Once this is in place, what does it make it easier for them? Yeah. I mean, one of the use cases, for example, of C1 is that we have an identity aware AI MCP gateway.

[00:21:36] So what that lets our customers do is deploy hundreds, if not thou, however many MCPs you have, including for internal data sources, wire that up into a unified control plane and then give access to data and systems and actions across your infrastructure to individual employees and agents that they're running, but in a secure way that's governed based on identity.

[00:22:05] Meaning I can still control for my SE team or my AE team or my SDR team, the data they can access, the actions they can execute across those downstream systems. We can govern that in runtime. We can evaluate behavior. We can make sure that the agents and the workloads and the actions that those agents are taking are staying within like guardrails.

[00:22:27] And so what that set of technologies does is it lets CISOs and CIOs move from you can't use any AI or, you know, making people kind of effectively work around the rules because like there's nothing in place to, hey, we have a governed path. It enables self-service. It's got security built into it by default. And we know that if you deploy agents and you're using this infrastructure,

[00:22:53] those agents are going to operate, you know, soundly and they're going to make good decisions. Or if they don't, we're going to, our infrastructure is going to prevent them from doing destructive actions or bad things. And so that's, you know, that's really the spear of the fearlessly in like scale AI fearlessly is we want to give customers really the confidence to be able to adopt AI without worrying about the implications of what it means for their business. And that means you got to have visibility.

[00:23:21] You got to have runtime security enforcement. You got to have self-service. You have to have governance. And that's, that's been landing super well with customers. I mean, they, they love the message. Employees love using C1 internally. It's been really exciting actually to see the market pick up on it. And a few moments ago, you said there's so many different companies seeking you out at the moment, look, where's they look for the solution. I'm curious when they first arrive at your front door, are there any trends in the kind of questions that they're asking?

[00:23:49] Is it, is it all very similar across different industries all coming to you asking the same question? Yeah. So it's definitely different by different industries. I would say, I think, you know, there's a set of kind of more tech forward, tech leaning or AI native businesses where they're adopting AI tools. They're really looking for like specific types of technologies. They're looking for our AI MCP gateway, or they're looking for a more automated governance.

[00:24:17] So they kind of know exactly what they need. And they're just partnering with C1 on, you know, different parts of the stack to help them accelerate their journey. But they're fundamentally sort of builders. And I would say there's a different set of customers where they, maybe they didn't have their identity program kind of dialed in, and they're still doing a lot of manual work. And they're seeing AI coming and accelerating everything.

[00:24:43] And they're realizing we really need to invest in this because we need the, we need the basic hygiene and set in a platform in place for us actually to accelerate our AI journey. And maybe AI is not affecting their, their business or their industry directly at this moment, but they see it coming and they know it's around the corner. And so they're, they're really trying to go through that identity transformation today so that they can accelerate into the AI future and transition themselves into agentic enterprise,

[00:25:12] maybe six months, 12 months down the road. So everyone's I, the way I like to talk about our customers is everyone's at a different part of the journey and that's okay. What we want to do though, is partner with customers on that agentic enterprise transformation, regardless of where you are, whether it's just automating manual processes more in the human world, uh, in the human identity world to really accelerating AI adoption at your business and deploying agents at scale and AI at scales,

[00:25:40] your employees will partner with any of those customers. Um, but we just want to be that strategic partner. That's going to help you accelerate your business through identity and agentic transformation. And for anyone listening, maybe we've ignited their curiosity or set off a few light bulb moments today and they're heading into the office. If they want to get in touch with you or your team or read more about all the announcements, everything we've talked about today and also keep up to speed with new things as they arrive throughout the year. Where would you like me to point everyone?

[00:26:11] I just go to C one dot AI tons of great content there. Great blog posts, great tutorials. We have a C one Academy, uh, that's got a set of videos there, uh, that talks about a lot of these concepts. Um, so we, we've really tried to create value in the ecosystem by just giving away a lot of, um, kind of knowledge and sharing as much best practices as we possibly can. So that's a good place to get started. Awesome. And as I said earlier,

[00:26:37] the enterprise now runs on humans and agents together and securing this takes one control plane for every identity. So again, I applaud you for everything that you're doing here. I love to stay in touch with you. Maybe get you on back on the show next year and see how things are continuously evolving. I will add links in the show notes to everything you mentioned. I encourage people listening to check you guys out, but just thank you for sharing your story today. Yeah. Thanks a lot, Neil. Thanks for having me. I appreciate it. Always happy to be back anytime.

[00:27:05] I think Alex left us with a useful way to think about AI and work. People may spend less time producing artifacts and more time building the factory, choosing inputs, directing agents, and reviewing outputs. And this is what multiplies productivity, but also the consequences of weak access controls. An AI agent needs an identity, a responsible human,

[00:27:35] limited permissions, protected credentials, and real time checks on whether each action supports its mission. And quarterly reviews cannot govern something taking thousands of actions in minutes. So I really appreciate this move from telling employees they cannot use AI to giving them a secure route without surrendering control. Again, massive point there. So a big thank you to Alex for joining me today. Remember,

[00:28:02] you can find C1's resources and academy over at C1.ai. And if an agent acted beyond its mission today inside your organization, are you confident you could identify it, track it to an identity and stop the next action? Food for thought. Let me know. TechTalksNetwork.com. Keep your experiences and what's worked for you and what hasn't coming over to me. And if you want to come on the show, also let me know.

[00:28:32] But that's it for today. So thank you for listening as always. Bye for now.