How much time passes between discovering a vulnerability and getting the right person to fix it? On Tech Talks Daily, I'm joined by Vineet Edupuganti, co-founder and CEO of Cogent Security, to talk about the work that takes place after a scanner finds a problem.
Vineet describes it as an assembly line: someone has to find the affected systems, work out where they're exposed, identify the owner, and get the fix completed. Each handoff gives a ticket somewhere to get stuck.

One of the simplest examples Vineet gives is a vulnerability sent to the wrong person. The agent may have identified a real problem, but the recipient can't act on it. The ticket goes back and forth while the system stays exposed. We talk about how Cogent brings together security and operational data to help agents determine who owns what, and why the quality of those records affects what you can ask an agent to do.
We also discuss what Vineet calls governed autonomy. He describes customers setting their own boundaries and starting in environments where the consequences of a mistake are easier to contain. A development system and a customer-facing production service may need different permissions. He makes the case for auditable actions and limits on what an agent can change. I ask where analysts still need to be involved, especially when a fix could interrupt the business or when someone has to weigh a security risk against the cost of downtime.
For anyone considering another security tool, Vineet suggests looking at measures such as time to remediate and how often teams fix issues within their agreed deadlines. We also discuss the trade-off between a broader platform and a specialist tool that may do one job better.
I'd love to hear what you're seeing in your own organization. Which handoff would you trust an agent to handle today, and where would you still want a person to make the call? Please share your thoughts.
Useful Links
Connect with Vineet Edupuganti
Learn more about Cogent

[00:00:00] Do you need AI agents that you can trust? Well, with an AI data layer providing real-time connection within your data platforms, you can trust your agents to provide accurate solutions. So, scale your business by trusting your agentic AI accurately getting the work done for you. Trust its capabilities with Denodo. And you can do that by simply visiting denodo.com to learn more.
[00:00:29] How long does it take your company to fix a vulnerability once somebody finds it? Before an engineer touches a patch, somebody has to work out where the software is running, whether it is exposed and who owns the system. Send that ticket to the wrong team and it can spend the afternoon getting passed around while the underlying problem stays exactly where it surfaced.
[00:00:53] Well, today I'm joined by the co-founder and CEO of a company called Cogent Security. And we're going to talk about where AI can take work out of that process and how companies decide what an agent should be allowed to change. He will explain why a team might let an agent act in a development environment while keeping approval in place for production.
[00:01:18] But we will also discuss how to tell whether a new security tool is actually helping you fix problems faster, especially when the scanner is already finding plenty more. But enough from me. Let me officially introduce you to him now. So, a massive warm welcome to the show. Can you tell everyone listening a little about who you are and what you do? Absolutely. Thank you for having me, Neil. My name is Vinit Edipuganti. I'm the co-founder and CEO of Cogent Security.
[00:01:47] We build AI agents that analyze, prioritize, and remediate security vulnerabilities. My background spans everything from machine learning engineering to working in the world of security. Prior to Cogent, I was at a company called Abnormal, which was a fast-growing email and collaboration security company. It's been about two years at Cogent. And since then, we've seen a tremendous change in the threat landscape.
[00:02:11] Mythos, obviously, earlier this year, followed by the OpenAI Hugging Face incident, followed by PHP. So, it's a really interesting time in the ecosystem. Risks levels are high. And I think one of the most important aspects is that life for attackers has gotten way, way easier than it has for defenders. And that's really the gap that Cogent is intended to go and bridge. Yeah, I think the risks have really been highlighted this year. And currently, as we record today's episode, there's been a huge outage at Microsoft there.
[00:02:40] And security teams, they face a growing volume of alerts and different products, etc. But I'm curious, what problem did you see that would lead to Cogent security? I think for me, it was really a combination of three factors. I think, number one, I'd seen the problem myself, right? I had worked with teams fixing vulnerabilities. I'd seen some of the challenges in the landscape. I think, number two, talking to customers. That's always the best way to really uncover pain and figure out what's important to people.
[00:03:09] So, I spoke to dozens and dozens of organizations before we even founded the company. They would all complain about vulnerability management and exploitation and the difficulties in keeping up. I think the third piece was just looking at some of these secular trends, right? And with the rise of coding agents and more and more infrastructure sprawling, the attack surface continues to grow. At the same time, hackers can feed one prompt into an LLM and exploit vulnerabilities way faster than before.
[00:03:35] And so, already, a couple years ago, vulnerability exploitation was the fastest growing cause of breaches. And it was pretty apparent that on the same trajectory, unless you had the same kind of technology deployed defensively to go and offset that, that was only going to continue. And I think so far, that's played out to be true. And to bring to life some of what you're seeing and hearing and the journey that you've been on, you don't have to name any names here, but is there an incident or a customer experience that you've seen
[00:04:01] that maybe shows why the existing or traditional security workflow is just no longer enough? I think the best example is actually probably the PHP to shell incident from a couple of weeks ago. Yeah. I think that was notable for a couple of reasons, one of which was just the fact that having that level of prominence of cybersecurity, right? Like front page of the Wall Street Journal, you know, that's not something that would have been common a couple of years ago. So, that was one thing that was really interesting. And I think the most impactful aspect, as you look at some of the challenges that organizations were facing,
[00:04:31] was that very first, they really struggled to even know where they were exposed. And so, they had all these instances of PHP running, but even knowing, hey, it's running here, I potentially could be exposed, I need to go fix this thing. Those were challenges that many organizations were not able to answer. I think the second piece they struggle with is, hey, I am exposed here, I know I want to fix it. Who even owns it? Where do I route this to? And then actually getting the fix completed. Every single step in the vulnerability management lifecycle, you can think of it as, you know,
[00:04:59] an assembly line with all these handoffs, there's friction that's involved with each of those steps and it takes time. On the flip side, you're being attacked by AI agents that don't sleep, they're relentless, they're trying every single path in parallel. And so, it goes to show until that assembly line becomes fully autonomous on the defensive side, the attackers will always have the upper hand. I'm curious, for people listening here, what business results should a company expect before adding just another security platform to each stack?
[00:05:27] I think generally when we talk to customers, there's three dimensions of value. I think number one, consolidating the existing stack. There's a lot of tools that are out there. And I think especially in this era of platformization, many companies are interested in taking what they have, eliminating some of that and really whittling down. So, that's one piece. I think number two is around risk reduction, right? That's the name of the game, the reason why we're all here in the first place. So, there's different ways to measure that.
[00:05:54] People might look at the mean time to remediate a vulnerability or how often they're fixing their vulnerabilities within SLA. So, looking at some of those risk-based metrics and seeing if they're improving upon testing a solution. And of course, that has second-order consequences, improves your compliance, reduces the probability of a breach and all the negative consequences that can come with that. And then I think the third, which is very concrete, is just efficiency. There's a lot of time that goes into running a lot of these security workflows, not just from security teams, but even from engineering teams and IT teams
[00:06:24] that oftentimes are the folks who have to actually remediate the vulnerabilities. And that becomes a very clear cost savings where maybe before you only needed or you would have needed a full 10 people to go and do something. Now you need five. This incremental five headcount can go be allocated towards other problems. So, that's certainly the way that we see customers thinking about it. And there are, as you said, a lot of tools out there. There's also a lot of hype and a lot of big promises being made by some.
[00:06:50] So, from what you're seeing here, where do integration, data quality or ownership problems, where do these things collectively make security automation very often unreliable? Yeah, I think that's the hidden reality of AI in general, right? Is that garbage in, garbage out. If you don't have good quality data, you're not going to have a great AI result. And the challenge is that security data, to your point, is very high scale, unstructured, messy, comes from a variety of sources. So, that's actually been one of the core innovations of Cogent.
[00:07:20] And one of the main areas that we've been investing in is really building this context graph that allows us to bring together disparate data sources, join it, contextualize, reconcile when there's conflicts, and then have a reliable base upon which AI agents can actually go and operate. And time and time again, we see customers that do not have that reliable data foundation, really struggling to get answers in a timely fashion. Perfect example of this in the wild, if I have a vulnerability I want to fix, and I send it to the wrong person, nothing's going to happen because that person doesn't care, they're not able to help.
[00:07:49] And now there's this ping pong back and forth happening until you finally do get it in the hands of the right person. And in that time span, so much of a gap exists where you're exposed to a significant amount of risk. So, you know, those are the types of consequences of not having that strong data foundation. So with all that in mind, I'm curious, how do you decide which security decisions can be automated and which will stay with a person? I know it's a delicate balancing act, but any tips or advice on that?
[00:08:17] We had a lot of philosophy on this and thoughts. And then I think ultimately where we've landed is it's not our decision to make, right? And so we think of it as governed autonomy. Our goal is to figure out what should happen by when, who should be involved, what's the risk level from a security standpoint, from an operational standpoint. We're going to bring all that together. And then we let the customer define those zones of autonomy and think of it like a self-driving car, right? Or maybe like a Tesla would be the best example, right? There's people who can drive the Tesla. There's people that can turn on full self-driving.
[00:08:46] So I think the same thing is true. And we see customers oftentimes starting with lower risk areas, could be workstations, could be their dev environments, and then slowly building up the comfort to move to prod. But everybody's going to have a different risk tolerance. Everybody's going to have different organizational constraints. And so we leave those decisions of where they want to automate up to them. So some teams will, listening, they will want a single platform. And others might prefer those technical teams, might prefer those specialist tools.
[00:09:15] But what are the real trade-offs here? You know, I think it really does come back to that same question of ROI. I think like all else being equal, the platform could be better, right? In the sense of one vendor relationship to manage. Things may talk to each other in a more seamless fashion. However, you also get vendor lock-in. And then oftentimes, right, product number seven from your platform solution is not going to be nearly as good as the actual point solution. So I think there's going to be a little bit of a rotation back where, you know, for factors
[00:09:44] including desire for cost consolidation, as well as just the bloat in the security stack, people have wanted to consolidate. But now I think at the age of AI, efficacy is the only thing that matters, right? Because if you have, you know, a perfect consolidated solution that's not actually solving the problem fully, that could lead to a breach. So I think more and more, we're now going to start to see this pivot back to best of breed and folks really just wanting the best hammer over the nails that they have. And we've already mentioned the hugging face incident, more recently the PHP issue.
[00:10:13] And I'm curious from everything you're seeing and hearing here, what are or what new risks appear when AI systems are given authority to investigate or respond to threats? Because we're now entering this phase where we're not just talking about humans and agents working together, agents working on their own, agents talking to other agents. Any risks you see here? Without a doubt. Right. I mean, agents can be wrong. They can make mistakes. There could be unintended consequences. So I think guard railing, how agents are used is super important.
[00:10:43] Limiting the actions they can take, having the right auditability, having deterministic guardrails. It's kind of like at a bowling alley when you have the bumpers on, right? That's maybe one mental model for thinking about how to keep these agents on the rails. But, you know, also a big part of this comes down to like risk assessment, right? And I think before it may have been that full autonomy was not something that made sense because the risk wasn't high enough to warrant it. And maybe you go and take a system down, you cause disruption. And that just is not worth potentially considering.
[00:11:14] However, now you still may have some of those risks. And of course, I think the right AI analysis can reduce them. But the reality is like the patch window has disappeared, right? And so if you're not resolving your security issues quickly, especially the high importance ones, I mean, again, with very high probability that could lead to an incident. So I think as those dynamics start to change, the ROI calculation of autonomy shifts and people are going to be more bullish and moving in that direction. I don't think it's an overnight shift. I think it will still take time like with anything.
[00:11:43] But it's hard to imagine a world 12 to 24 months from now where full autonomy of many security workflows is not the norm. And we have talked about some of the worst case scenarios here, but there is a lot to be optimistic about. So looking to the future with a more positive, proactive approach to security mindset, how do you see automation reducing pressure on analysts and some of that well-documented alert fatigue that we've heard about over the last few years, but without removing the context
[00:12:11] and the judgment that makes their work so valuable? Yeah, I think like the future world that we're kind of envisioning here, right, is one where we're going to be attacked by AI swarms, but we have defensive AI swarms that can go and combat them. A lot of that low level analysis and activity can be handled by the agents. And then humans are involved for the more strategic aspects, right? Architecture decisions, policies, feeding context, right? They always become like the managers of the AI agents.
[00:12:38] I think that's going to be the predominant mode of operation. And I think that's also desirable because oftentimes, you know, looking through millions of data points and trying to come up with an answer, it's not something that humans are particularly well-suited for, nor do they necessarily enjoy. And so I think that bifurcation is actually the right one for the industry. And after doing a little research on you, I know that your mission is to prevent breaches using
[00:13:01] autonomous AI systems that continually secure systems against AI-armed adversaries that are operating at machine speed. But for all you're working on here, what excites you at the moment? Yeah, I think really it's just number one, the mission, right? I think this is something that has to exist in the world. And if you think about not having this technology, companies across the globe are going to be in a state of constant breach, right? Which just erodes society basically. So I think the impact of the work we're doing is very, very significant. I think we're on the
[00:13:31] frontier. One very interesting thing about cybersecurity is it's an application where you can't be wrong because if you're wrong in one direction, the customer gets breached. If you're wrong in the other direction, potentially you could cause significant business disruption. So it's not just like toss tokens in and get them out. It's can you actually thoughtfully design a solution to a problem that really is able to address the challenges customers have and do so reliably. I think that's very interesting. And I think the third thing is, you know, getting customers
[00:13:56] comfortable with this new paradigm. It's a UX problem. It's a trust problem. But our goal is to really be that guide that can take them from point A of the status quo into that new future that inevitably has to exist and what the operating model is for security. I think we've already had some initial success with that and want to continue pressing our work on that front too. And we are recording this at the end of 2026. And again, doing a little research,
[00:14:20] I saw that you've also been included as a sample vendor in Gartner's top strategic tech trends for 2027 under autonomous exposure remediation. Feels like a big moment. Tell me more about that. Yeah, I mean, I think the recognition is great. I think, first of all, at the category level, right? If you look at what Gartner is saying are the top tech trends in general for 2027, to have cybersecurity related aspects and specifically what we're solving in the top of the list, I think is a testament to just how important it is. And also how, you know,
[00:14:50] the elevation has occurred in terms of the C-suite. It used to be that these problems were things that maybe a director and the chief information security officer might care about. Now it's the CIO, it's the CEO, it's board members, right? So I think there's a lot of prominence and visibility here. And folks realize that solving or exposure management and remediation is one of the ways that they keep their business up and running and also advance it as we go into the next couple of years. And then of course, us being featured is a subcomponent of that.
[00:15:17] I think obviously great testament to the work our team has done. I'm very grateful for the customers that have chosen to partner with us. The best is yet to come. We're just getting started, but I think it's all, it's all positive in terms of indicating what the potential is here. It certainly feels like there's exciting times ahead for you guys, all that hard work paying off. And I suspect you can't reveal too much, but is there anything you can reveal about your focus for 2027? Now it is just literally weeks away. Anything you can share about what you're going
[00:15:45] to be working on, what you're going to be focusing on next year and any teasers you can leave us with before we get you back on next year? Yeah. I mean, I think the way that we've been describing ourselves as a cyber defense factory for the enterprise and what does that mean? It's AI agents that find and fix vulnerabilities before attackers can exploit them. And so continuing to push on all the different dimensions that are required to make that true, that is really our core focus over the next 12 months. And that involves everything from going deeper on remediation and handling some of the advanced edge cases that come up
[00:16:15] in complex environments, especially for big companies, deeper work on discovery. We made an acquisition earlier this week that pushes the envelope in terms of our capabilities in that regard. So very exciting, lots of things to do. We're growing our team rapidly on both the R&D side and also the go-to-market side to go and seize the opportunity and deliver for customers. Well, I love chatting with you and I think we will have to get you back on later next year, find out how the landscape is evolving. I'm sure there is lots more to come, but for anyone listening
[00:16:44] wanting to dig a little bit deeper on anything we talked about or learn more about the work that you're doing, how you might be able to help other businesses, where can they find you online? You can find me on LinkedIn. You can find Cogent on LinkedIn or on Twitter as well. Website is Cogent.com. Any of the above are great ways to get in touch and look forward to hearing from you. Awesome. Well, I'll include links to everything, including your LinkedIn and indeed the company LinkedIn page as well, so people can follow you and learn more about those announcements
[00:17:13] as and when they arrive. So many big talking points here. I'd love to throw it out to people listening, what they're experiencing, what today's conversation meant to them. But more than anything, just thank you for bringing all this to life today. Really appreciate you, Tom. Appreciate you having me, Neil. Thank you. So many great points in our conversation and particularly the one that vulnerability or the risks that vulnerability when they're sent to the wrong person. Everyone in that chain might be doing
[00:17:40] their job, but nobody can fix it until the request reaches the person who owns the system. So maybe start there before asking an agent to take on the whole security workflow. If you lead a security or engineering team, pick one recent fix and follow it from the moment the issue was identified. Find out where all the time went. If it was spent chasing an owner, improve the records that connect software to teams. And if the delay came from deciding whether a change could
[00:18:09] interrupt a service, agree who gets to make that call and what evidence that they need. So I've included links to Cogent's website in the show notes. Please share your thoughts. Where does your team lose the most time between finding a vulnerability and fixing it? I suspect you've all got a few war stories there. You've earned your stripes in the cybersecurity battlefield and I want to hear
[00:18:35] from you. So techtalksnetwork.com. Send me an audio message, send me a DM and I'll get back to you. But that's it for today. So thanks for listening as always. Bye for now.

