AI Agent Security: How to Stop Autonomous AI from Becoming Your Biggest Insider Threat

Aaron Crow (0:1.182) Thank you for joining me for another episode of the Protect It All Podcast. Man, I David and I were just talking about this before. and I I this is a theme I keep coming back to. I I get the privilege to sit across from folks, but my schedule's crazy. The guest schedules are crazy. So sometimes this has been two, three months coming of reschedules or putting things on calendars because at the end of the day, this is not something that is it it can't trump.

Aaron Crow (0:27.476) client work or or our priorities. So many times it gets pushed back. So with that, David, thank you very much for being here today. Thank you for being flexible with schedules and and rebooking and and being patient with me and getting back on. So I'm excited to to have you here today and and dig into your story and and all that is that you are all the things that you've got working on the table. So why don't you introduce yourself, tell us who you are, tell the audience all the things that they need to know.

david girvin (0:54.710) Yeah, I mean, dude, the pleasure's all mine. I'm I'm excited to be here. It's it's fun, get to hang out with cool people and chat. who am I? my name is David Gervin. my day job, I'm the lead technical advocate, security I AI researcher, whatever. I don't even know what my title is anymore, but at Sumo Logic, I do the things that no one else wants to do. I'm also the founder of Asuri, which is we were the first

david girvin (1:24.332) governance layer for agents. we were up and running last year actually. And then yeah, my background's weird. To put it bluntly. so I was a ship captain, for on and off for I guess 15 years or so, burnt out being at sea too much. And then I owned a welding and fabrication solutions engineering company in Hawaii. And

Aaron Crow (1:39.028) Okay.

david girvin (1:54.550) I broke my back, couldn't do the work anymore. And one of my buddies on my jujitsu team, I've been training jujitsu for many years, was Jeremiah Grossman, who started White Hat Security back in the day. And he'd always like hint, like, dude, you should be a hacker. And I was like, You're a nerd, bro. Like, what? I'm a blue collar, like I'm not a hacker. And

david girvin (2:20.343) Well, wh I when I was pretty much bedridden for a year, I could I became a hacker. Actually I I tried to become a software engineer. And and then I went to my first conference and they had Jim Manaco, had secure coding with Manaco, you know, thing and I jumped in there and I was like, this is okay. And then they had a CTF at the end and I just crushed it and then I figured out how to steal everyone else's points in the CTF and break the

david girvin (2:49.399) the the app and and then I became a security guy for one password after that.

Aaron Crow (2:56.935) That is freaking awesome. I have to say, out of all like I've I've had, you know, hundreds of guests and and episodes, it's definitely the first time I've had somebody that was a ship captain and then a welder, and then you broke your back and your jiu-jitsu guy told you to so I I train jujitsu. My my son and kids all do. My my son's a bluebelt, he's he's seventeen, he teaches at the jujitsu gym and possibly opening up one anyway. So jujitsu's big in our family as well. So

Aaron Crow (3:24.357) that that makes a lot of sense honestly when you say it and and that's the what drove your your path to the next step that you were gonna go to.

david girvin (3:32.055) Yeah, it's nerds, bro. Like jujitsu people are nerds, hackers are nerds. It's just a bad it's a bad place to hang out. yeah, and I I still I'm usually one of the guest coaches at the BJJ SmackDown at Black Hat every year. So that's been like super fun to keep it kinda, you know, keep the nerd connection going from hackers and and jujitsu guys. So we we we drag Forrest Griffin every year into it and and make him talk about tech. It's pretty hilarious.

Aaron Crow (3:34.237) Yep.

Aaron Crow (3:36.904) Yep.

Aaron Crow (4:0.498) Yeah. Well, the funny thing is, is I've been going to Black Hat for I don't know, 10 plus years. And it wasn't until last year that my one of my professors was there and and he goes to Black Hat, he's a cyber guy as well. and afterwards he's like, Why weren't you at the jiu-jitsu thing? And I'm like, What the hell are you talking about? And he's like, You didn't know there was a jujitsu thing black hat? I'm like, No, why didn't you tell me, man?

david girvin (4:7.928) Mm-hmm.

david girvin (4:27.702) Yeah, it's been a good time. If if you guys are ever going, Jeremiah puts it on and it's usually we usually get a couple of like sponsors or whatever, but it's just go hang out. Like we have a bunch of newbies there. I'm usually the newbie coach because people find me amusing. I don't know. Do I f do I look safe? I feel like that's why they put me with all of the new people. But and then we have food trucks. So everybody gets to hang out and chat. But it's always a good time. There's everybody there from people who are like

Aaron Crow (4:44.680) Ha ha ha.

Aaron Crow (4:53.342) Sounds awesome.

david girvin (4:55.170) Geez, almost coral belts, you know, tons of black belts all the way through white belts or first day. So yeah, if you're coming, if you're going to Black Hat, find us. It's always fun.

Aaron Crow (5:0.456) Yeah, yeah. That's awesome. That

Aaron Crow (5:4.402) That that's that's the life of jujitsu in a good place that you go to, right? It's it's that camaraderie and and team building and and getting outside your comfort zone and and all those things that are just all wrapped into one.

david girvin (5:17.068) Yeah. Well, and like we work a stupid job. It's like it's horrible for your health, it's horrible for your mental. It's like you gotta find some sort of healthy outlet, you know. I've I was an athlete my whole life, so that's one thing. All the guys that have ever been on my team, I force everybody to do something that's not computer stuff.

Aaron Crow (5:26.355) Yeah.

Aaron Crow (5:32.615) Yeah. Yeah. Absolutely. Well, that's awesome, man. So so you you had this amazing background. And and the funny thing is, is is yes, I've never had anybody that's been a ship captain, but I I talk a lot about as people come to their career. Most of the people that are sitting across from me and I'm talking with, they're they're it's never, very rarely is it a straight line trajectory of

Aaron Crow (5:57.011) You know, I knew I wanted to be this and I I went to college for that. And then I have that job and I've been in that career my whole life. You know, usually it's like ping pong and it's all over the place and and you're moving around. But every one of those things is experiences that you bring in. and that's a perspective that you know, I'm sure there's a lot of things that being a ship captain helps you be a hacker. I'm sure there's a lot of things that being a ship captain helped you be a welder and welder to be a a a hacker, all of those things.

Aaron Crow (6:26.482) They all are are experiences that you bring into the the field that you're at and what you're doing and the seat you're in.

david girvin (6:32.916) Yeah, I mean a hundred percent. I mean, especially with soft skills, you know, management, leadership. you know, start I mean, I started my own welding company, you know, so it's startup culture, you're s you're hustling, you gotta learn sales, you gotta learn all that kind of stuff. And yeah, I really kind of noticed a lack in a lot of that when I first started. so I was at One Password and I worked for a couple of small startups and then I was I built the Appsec side of Red Canary.

Aaron Crow (6:45.449) Yeah.

david girvin (7:1.982) And worked with Bit Discovery and a few other ones on my way to sumo. And so I mentored a lot of people into the space, which was amusing to me because like by the second year I was in security, I was like, yes, let me be your mentor, which is asinine at its best. but the one thing that I I did it very differently because I always took a different approach. Like for me, AppSec, when I was doing that side of the house, it was always inside sales.

david girvin (7:26.818) So I treated engineering like a sales guy. Like I took people out for drinks and you know, wine and dine, built champions, all that kind of stuff. So I think that just rubbed off on whoever was working with me at Mentor. I used to make people go to Toastmasters, which is what all the real estate bros go to. Yeah. So like get these guys who are brilliant, but they're just real introverted. And they're like, You want me to do what? And I'm like, I won't work with you unless you go do this. Because

Aaron Crow (7:41.128) Yep. Yep.

david girvin (7:53.069) You know, anybody can get in, but individual contributors, they don't really move up as easily as someone that has those soft skills that can show some sort of leadership or program level management type of thing. So I wanted people to be able to sell themselves 'cause that from the beginning that was always like a blocker I saw.

Aaron Crow (8:9.736) Yeah. Yeah. It's it's funny because throughout my career in the beginning, it was all technical. It was, you know, am I the smartest guy at this firewall or this capability or whatever it is? And I was with a a company and they had this thing they called leadership circle. they brought in an outside consultant, and that's basically what we did. We did book reports and public speaking and business use cases. And and in the beginning I was like, This is stupid. I'm not in college anymore. Why am I doing this?

Aaron Crow (8:38.630) I don't want to freaking do this. And it was all in addition to my normal job. and then I did the first year of it, and then I got promoted to do the second year, which was like the more advanced one. And then I got to the third, third year, which was like the elite. There was only like five of us in there. And looking back, it's the thing that catapulted my career because of those soft skills. Like they were buried, but I never used them. I never, you know, really exercised those muscles.

david girvin (8:42.370) Mm-hmm.

Aaron Crow (9:4.908) And those are the things that were value add to my career and got me out of I still do hands-on, I still do technical, I'm still coding and AIing and all the things, but those other software are the things that made me more valuable to my customers, valuable to my employer, valuable to, you know, things that I'm doing for myself. All those things were more better with my wife and my kids. Like all the things were better all the way around.

david girvin (9:30.859) Yeah, I agree. And it's it gets gives you a lot of options, right? Like, you know, because I kind of had that repertoire built out, it was a lot easier for me to start doing conference talks, which really helped my career. I mean, I've done like, I don't know, forty something call for papers now, I think. And you know, there's just a little bit of career credibility. I feel like when you can really put together something like that that's good and has good research and helps other people with a problem.

david girvin (9:59.639) The other side of it is I think a lot of security people especially think of like, like success for me is like staying in this highly, you know, technical position, maybe CISO. Man, you can be like, I'm probably not gonna be the smartest guy in the room in security unless we're talking about like agent governance. But outside of that, like it's not gonna be me. But on the marketing team, I'm killing it. Like

david girvin (10:28.758) You know what I mean? Like that's the funniest thing. I never thought I would roll up to a CMO, but I've got like a really brilliant CMO at Sumo in my day job. And they just kinda give me free reign. So I've built out like the our agentic strategy for marketing is probably much deeper than ninety percent of engineering teams. And like I wouldn't have ever had that opportunity if I didn't do front of house stuff. I didn't, you know

Aaron Crow (10:29.598) Ha ha ha.

david girvin (10:57.602) Do those talks, do some sales stuff, field CSO world. And it's like my my day job's killer. Like, you know, it's like it's it's good. It's not stress. I'm not worried about waking up at three AM because we have a breach or anything. And I still do security every day. So yeah.

Aaron Crow (11:9.714) Right. Yep. No, that that's awesome. And and all of those things. So, you know, I I do stuff like that. I've I've got a friend that has a physical security company and you know, he needs help with digital protection for celebrities. Like that was a a weird one. And then I've got another buddy that does, you know, marketing. and he he reached out and needed help with, you know, marketing for sports teams and helping them sell tickets and get sponsorships and

david girvin (11:24.238) Mm.

Aaron Crow (11:37.883) All the things I've done. So I've ran events. So I've done cyber events and and you know, I do them in I live in Texas. So I do them in Texas and you know, they're all shooting and you know, bring CISOs out, put a gun in their hands, put a helicopter, shoot out of a helicopter, end of the day with bourbon and cigars, like it's it's a it's a fun day. but you know, all of those things together, like you see the day you had a great time, but that took months of time.

david girvin (11:48.067) Yeah.

david girvin (11:56.374) It's a great.

Aaron Crow (12:4.008) To put together and work to get sponsors and vendors and all and t-shirts and swag and all of those things that go together to do all those things. None of those things are cyber related, but they all help to get to the end point that was cyber. Like the the final the the the final piece of the puzzle was was cyber related. The rest of the things was marketing and sales and people, people and all that type of stuff. And and it's fun to do, but those skills are not something that

Aaron Crow (12:33.160) you know, you have naturally by just going and getting a certification and getting your C I S S P and it it's it's the thing that differentiates you from the from the crowd as well though.

david girvin (12:38.881) Yeah.

david girvin (12:43.276) Yeah, I kind of systematically approach that part of the house. Like I knew I needed time and go to market to understand the space 'cause I always thought I'd kind of run my own thing. That's my my buddies do that. And and then, you know, I need to learn marketing and SEO and all you know, like how do you put on events? Like w how do you get a CISO to talk to you about stuff, which is a nightmare, honestly. But the funny thing you just mentioned is, dude, shooting the I swear the deeper you get into security you realize like in the in the in the

david girvin (13:13.186) Real low levels, like practitioner level, like, shooting, weird. I don't think that's you get into the CISO or the deep technical guys, like I don't want to just drop in names, but people trip all the guys that I shoot. And then like I that was the funny thing 'cause I I I grew up in Texas too, so I grew up shooting it. But I've taken so many of our European either customers or employees out shooting anytime we're in Vegas, I'm like, You guys wanna go shoot machine guns? They're like

Aaron Crow (13:27.346) Yeah.

Aaron Crow (13:37.832) Mm-hmm.

david girvin (13:43.025) can we do that? I'm like, yes, we can. And everybody has the best time in the world. You know, it's it's just funny. I feel like that's like our industry's hidden little secret, is the deeper you get, everyone shoots.

Aaron Crow (13:44.924) Yes you can.

Aaron Crow (13:54.725) That's right. Well, and it's, you know, back in the day it was, you know, golf and and I'm that's still a thing, obviously. But I I'd much rather spend an hour at the range or two hours at the range that at that at a gun range than it than at a golf range. you know, it it's amazing. The the entire event and and and and obviously we could pivot after this, after this, but my entire idea behind everything with that was never to be selling. Like so I would bring in a sponsor and say, Hey, you're not gonna have a pitch deck.

Aaron Crow (14:23.486) You're not going to have a presentation. You're going to be paired with other CISOs and y'all are going to talk. You're going to hang out. You're going to eat lunch together. You're going to shoot. You're going to have a good time. You're going to drink some bourbon. And at the end of the day, you've made a relationship or you haven't. That's all on you. But at the end of the day, the sales come next week and the week after that. When you've built that relationship, don't pitch today. Cause if you pitch today, you're going to push them away. They're not going to want to talk to you. They're going to go hang out with somebody else.

david girvin (14:44.332) Yeah, yeah, yeah.

Aaron Crow (14:53.128) that isn't trying to pitch because they get pitched to your point all the time. They don't need that.

david girvin (14:57.794) Yeah. I I'm I'm definitely paying getting paid back for all the times I ignored vendors when I was the the buyer. Now that I'm the founder, it's like my life has just turned into hell. I'm like, of course they told me to fuck off. Like that's just what's gonna happen every time 'cause you did the same thing to people back then. It's horrible. But but just to circle back before I keep going, I don't know, maybe it's 'cause I grew up I mean I played sports and stuff, but like I'm a surfer. Like that's my main thing.

Aaron Crow (15:7.891) Yep. Yep.

david girvin (15:25.656) But like shooting and car events are the way to my heart and I feel like to many CISO's hearts. Like I know more CISOs who like shooting, cars and watches. That those are the three things I'm always like, Are you into any of these? Or bikes. It's it's funny. We like kind of start herding on certain hobbies to try to get away from all this bullshit that we deal with every day.

Aaron Crow (15:31.848) That's right.

Aaron Crow (15:37.715) Yeah.

Aaron Crow (15:40.274) Yeah. Right. Yeah.

Aaron Crow (15:47.315) That's right. It it it's it's a clear way to kind of filter out the riff raff, let's put it that way, right? So and it it's really it's really hard to have your pitch deck when you're on a at a gun range, because it it's just loud and no, we're not we're we don't have time or space for that. So it anyway, so back to you the the the regularly scheduled programming here. Like, dude, and that is fine.

david girvin (15:53.324) Yeah. Yeah.

david girvin (16:4.638) Yeah.

david girvin (16:10.286) I'm gonna take you on on rant. Just go with it.

Aaron Crow (16:14.534) I am perfectly good with that. And that is where I live in this space, man. So, you know, it it's it's just to me, it's awesome because all these things matter. Like the reason I do this podcast, it's not because I'm I I don't pretend to be the smartest cyber person. I'm not the best engineer. I'm not any of these things. I've got a great career. I've been doing this for 30 years. I've done a lot of cool things. I've I've I've made a lot of good connections. You know, I I I ran a real estate office for I had my real estate license since 2006. I've

Aaron Crow (16:41.874) I did it, I did bank owned foreclosures in Texas during 2008 when all of that shit happened. I would get a hundred properties a month, evicting people out of houses. I've done all of this stuff and and all of those things, kind of like what I was talking about with you. All of those things help me be a better all everything, right? It helps me be a better cyber person, helps me be a better business person. The business acumen and and all of those things that come along, they don't teach you this stuff. Like you can't go learn this. Yeah, you can go get your MBA.

david girvin (16:47.683) Yeah.

Aaron Crow (17:10.300) Or you can actually go do business and actually learn how real shit works, right? You you could go take a marketing class, or you can go talk to the marketing folks and then you you make some changes and you see what works. Like don't go read about it. Go implement something and see what works. Like build a website, get get some funnels, get a have have a CTA, like have understand who your do you understand who your, you know, ideal customer profile is? Like what is an ICP? Like these are all marketing terms that I'm spilling out. What does that have to do with cybersecurity? Everything. Like

Aaron Crow (17:40.293) I I'm in I'm in services. I have to sell my services to someone, which means I have to get them to know that I exist and that I have this service available and get them and interested to talk to me to buy my service or my product. I was a CTO of a software company. Same thing. Like, yes, I I had some technical need around the product and the engineering, but most I would even argue more more of my job was focused on.

Aaron Crow (18:6.248) go to market strategy and getting people and being in the in the room and making sure they understood what we were doing and what's on the roadmap and that we were going to be able to resolve the things that they were interested in or cared about and all that kind of stuff that really had nothing to do with the T in the CTO.

david girvin (18:21.612) Yeah. I mean, that's just it's kind of the game now. Like you're you unless you're at a company that doesn't sell anything, your security job is not just security. It's always gonna be to be a technical expert for your go-to market team, you know? And that's especially the higher up you go in your career. It's just it was a na it was a very natural move to go to more like field CISO role for a while there. I I hated having a number tied to my head, so

david girvin (18:49.602) So I started a company which is even stupider, but you know it's like Yeah, it's called my mortgage. Yeah. It's called a a mortgage and five kids. That's my number tied in my head now. but yeah.

Aaron Crow (18:52.286) Right. You still have a number tied to your head. You're just putting it on your own head. Exactly.

Aaron Crow (19:3.016) So, so obviously AI is huge. It's everywhere. You know, you go to RSA, you go to Black Hat, you go to any of these conferences, every fucking vendor product, everything has AI tied to it. Most people don't know what the hell they're doing with it. Most users don't know how to use it. you know, so obviously it's a great place to be. And and and I love the idea, you know, we talk about I I was a also a big four consultant at at one of the big four consulting firms, right?

david girvin (19:13.635) Yeah.

Aaron Crow (19:29.970) And we talk about governance and programs and people process and technology. And a lot of times it's just this fucking document that that a that an expensive consultant like me helps you write and you put it into the policy and it it gets filed somewhere. And is it followed or not? Is it is it, you know, whatever? Like it it doesn't mean much unless it's actually used. It's like, you know, I I have a knife. Well, what do I do with this knife? Am I gonna cut somebody with it? Am I gonna open a package with it? Like

Aaron Crow (19:59.239) It's how I wield it and what I do with it that matters. It's the same thing with governance, right? So AI has the opportunity, just like anything, to be really, really powerful. But if it's not handled well, it can tear your world down because not because it's bad, but because you didn't tell it what you wanted it to do and give it guardrails to stop it from doing something stupid. It's like the, it's like your five year old. If you if I gave my five year old this knife, they could do some really stupid shit with it.

Aaron Crow (20:27.794) Right. They could also have s they could do some cool shit with it too though.

david girvin (20:30.818) Yeah. I mean, yeah, I mean that's that's the thing. And I think it's it's funny. I think the term just kind of got muddied. So like w the space I've been working in is like governing autonomy, governing action in agents. Like that's that's where my company's built. That's where a of my research like I kind of started, I was like I think there was like three or four of us all at the same time that hacked ChatGPT three when it came out. And we all like post on Twitter and then I sort like we all did like conference talks. I think my

david girvin (21:0.642) first one was at RVA sec that year. And like s then agents came out and are like, software is making decisions now. So like what are the attack vectors there? And then you know it got deeper with MCP and everything else. And so I kind of been in the space already before I started getting it. But man, the marketing machine with the muddying of terms and everything has been so crazy. And we don't even know how to talk about the space anymore. But like that's something I've struggled with because people are like, you're in the

david girvin (21:30.550) a agent governance space. So do you like automatically write policy for companies? I'm like, no, not at all. We literally control the execution layer for agents. And they're like, what? Because depending on where you come from, like in AppSec governance would be controlling execution. In GRC governance is that paper that you paid a big four for and you've shoved into a Google Doc somewhere and you've never looked at again. So yeah. It's but the reason I actually I

Aaron Crow (21:55.156) Yeah.

david girvin (22:0.107) A little over a year ago, about a year and a half ago, I was building a one of one of the first AI native GRC platforms. Like we were running models and doing all the kind of stuff and building vector databases and we thought we were super cutting edge. And and then I had an agent delete my prod database. And this is 14 months ago, something like that. And then it then it lied about it.

david girvin (22:28.694) And and then I just literally after I done punching walls and crying 'cause it was like two in the morning or something, I was like, Okay, I'm gonna fix this problem. And that's how I got into the agent governance space and that's when we started a Shuri. We were we were a fair bit ahead of the market, which was interesting. But yeah, you know how it is. It's like when you're trying to build things and and solve problems, you you wanna solve problems that actually matter. So

Aaron Crow (22:53.736) Yeah. Well, it's to use an analogy, I I worked on my car and I'm I'm changing the the the front suspension. I have to change the the the the actual shock and it's in a a new Tahoe which I just got rid of. So anyways, because really because of this issue, but it's an airbag, right? And and the the air inlet is right on the top. and

david girvin (23:16.408) Mm-hmm.

Aaron Crow (23:21.608) There is no human way that you can get to that. You can take it off, but you can't get it back in. So when I re put put the new shock in, there was no way that I could get a tool that could put that on and it wouldn't turn by hand. So what did I do? I went and found a wrench and I'm like, okay, what is my problem here? And I like got a torch and bent the damn wrench so that it would fit into this little cubby and I could turn it like, you know, an eighth of a

Aaron Crow (23:49.435) Centimeter every s you know turn. And it took me an hour and a half to tighten it, but I was able to do it. But that was the only way it was going to happen. I'm sure GM probably has a specialty tool that I don't have, but you had to I've had to find it. And it was because of that specific issue. I had a problem that I couldn't find a solution for. And it was either sit there and cry in my st in my street, you know, because I can't fix it, or find a way, right? And that that's where

david girvin (24:1.910) Yeah.

Aaron Crow (24:18.632) You know, most great products come from because you saw a need and there wasn't like you're looking around and nobody has the solution. Well, I guess I'll just build the solution.

david girvin (24:27.958) Yeah. I mean that's exactly what it is. But the interesting thing is kind of how the industry converged. So like obviously like my background's red team, like offensive security stuff. my co-founder, his background is like he built infrastructure for Raytheon and DOD and like, you know, he's crazy. But so I came at it as like I look at agents as like insider threats.

david girvin (24:52.534) You know, they kinda listen to you until you get about 75% context window filled, and they start to do really dumb crap, trying to fill in the space. And then obviously obviously they're more attack surface and a and a and a neat way to break things with poisoning and everything. But I think everybody else thought of them the way they think about people. So like a lot of the research I did, like

david girvin (25:16.576) One of them was proving that just in time tokens don't work with agents. So everybody started trying to solve this at the identity layer. We we like I I mean probably just a lucky guess, but we thought we had to do it at the execution layer. And then, you know, you saw all the big vendors come out with just in time tokens for non-human identities. And I was like, you know, I don't I I didn't take VC money, so I'm like, everybody, you're doing it wrong. This is why you can't have a dynamic just in time token to a dynamic session. There's always gonna be this giant gap on the end of it. And

Aaron Crow (25:22.932) Mm-hmm.

Aaron Crow (25:43.700) Mm-hmm. Yep.

david girvin (25:47.127) And then people, I think, recently started figuring out, crap, we got to do this at the tool call level. And you know, we've been functioning since November. So like we were really kind of far ahead. I know this is such a small time frame. We're like, that's like six, seven months ago or whatever, but in AI that feels like seven years. And with the it's been interesting to watch everybody. I can tell who was on blue team and who was kind of more red team on how they've been building.

david girvin (26:16.460) Like you can tell who's been in regulated environments because they're trying to stay fully deterministic. Like we we stayed, that was like one of our main things. Like agents aren't allowed to have credentials, so we had to do credential starvation. Fully deterministic. It has to live in the execution path. And then kind of everybody kind of has their own take. And like you can tell who definitely was like

david girvin (26:39.110) Not in the regulated environment, they're like, AI can govern AI. And I'm Why didn't it doesn't work? I hate to break this to anybody, but if you've been in security, you know probabilistic things. They're like, We have a 92% accuracy rate. Like, that's one out of 10 sucks. but everything's just kind of starting to converge right now, but it's so hard to get through all the marketing noise. So I think we're gonna see a big shift in the next six months. Like, people are really starting to understand the problem.

Aaron Crow (26:45.051) No.

david girvin (27:6.506) Even a little guy like me, who's like not a big money guy, but has a lot of the research, like all the major dang, what are they called? analyst firms, you know, have booked time with me recently, which is cool. And like we're getting to kind of explain that side of the house. But the reality is, is like it's exactly what you said. Like, we're two engineers that saw a big problem and wanted to build it.

Aaron Crow (27:18.643) Mm-hmm.

david girvin (27:33.411) But I didn't realize how crazy the industry is when you're on the founder side. Like, because as an engineer, you're like, I just want to fix a problem and and like all security tools kind of suck. They don't all suck, but you know how you feel when you use them. I'm like, I'm the security guy. I'm gonna make a tool that doesn't suck. And then you like run into like what the real real life is. It's like, no, no, they don't care if the tool works. They wanna pump you up with $20 million and go to market and like you go to a VC and you're like,

david girvin (28:2.360) Can I have a million bucks? And they're like, No, but we'll give you twenty million and you're like, What? I don't want twenty million dollars And then you start like kinda the industry's not what we thought it was for as a practitioner. So it's been a humbling ride for sure.

Aaron Crow (28:6.332) Right. Yeah.

Aaron Crow (28:17.342) You know, and and that's something I know we we talked about it before we started recording, but you know, that that's this the the path of you know it's really easy to say, I want to be an entrepreneur. I wanna, you know, especially right now, everybody wants to be an entrepreneur. What does that mean? we laughed because, you know, when I I I led with, you know, I woke up at three o'clock in the morning and couldn't back to sleep. So I'll just get up and work, right? And and that that's literally been my day. And it it happens a lot.

Aaron Crow (28:47.002) and it it's because, you know, I am an entrepreneur, I am running a business, I am, you know, I have a nine to five, I I have a podcast, I've got, you know, my hand in all these different things that I'm doing. and I I I want that. I I I signed up for these things. It's not like somebody's forcing me and I'm not like a held against my will or anything, but you know, it's the thing that drives me. you know, speaking of like I've been doing jujitsu for 10 years.

Aaron Crow (29:16.612) and and I'm ashamed to say that I'm still a white belt because I don't train as much as I should, or obviously very often at all. and my my my coach obviously is giving me a lot of shit about that constantly, especially since my seventeen year old son just got his blue belt. and and and he's also six four. So yeah, I'm I can still I'm still stronger than him, but yeah, it's probably not long, that's not gonna matter anymore. So

david girvin (29:42.925) Nope. The reckoning is coming. Yeah, you both I mean, we both just have really bad life choices. like to me, I've always been like kind of that independent entrepreneurial spirit. It's such a stupid choice sometimes. It's just like, man, it's so much easier if you can be if you can have a good day job and just be happy in it and then like cut work off. And like you just lose, like we were saying, like.

Aaron Crow (29:47.527) I know, I know.

Aaron Crow (29:51.112) That's right.

Aaron Crow (30:0.639) Yep.

david girvin (30:11.502) For a year and a half I've been working two full time jobs and having kids and working eighty hours a week, you know? It's you wake up at five, you start taking mornings, you know, meetings in Europe and then you get your stuff done, then you go into your day job and you work all day there and you get off and you kiss your kids and then you work until, you know, twelve or what at ten or and that's just the life. And it's like I think it's I think when you really kind of believe in what you're doing, it it kind of works.

david girvin (30:39.982) But at the same time when you see how stacked the industry is against

david girvin (30:46.744) Trying to do the right thing, it it starts to wear you out. And like, not to completely black my ball myself from security, but I don't know if you guys know, but the way the world works now is you're kind of forced into you don't you're not forced. There's there's a chance you can bootstrap, but we've we've stayed no raise yet. but you do kind of get forced into this VC private equity game.

david girvin (31:13.654) And I, if you've worked in that space, it's pretty brutal. And it's not about the best tool that wins. It's about pumping up go to market and using their networks to kind of force software through them. It's kind of like a big Ponzi scheme a bit. And it's a little bit demoralizing when you've been like a security guy for 10 years or 30 years and you're like, I just want to like make shit safe. Like, is that so bad? But yeah, I don't know. It's it's a tricky space. Like

Aaron Crow (31:34.196) Mm-hmm.

david girvin (31:43.329) I'm probably the most proud of what my partner and have and I have done with Ashuri out of all the stuff I've done. And I've been to you know almost 50 countries and traveled all over the place, besides like marrying my wife, because she's rock star. Like if I could social engineer that woman into marrying me, I'm the best security guy in the world because I'm dirtbag, she's amazing. But besides that, like building like something that actually works and really helps people and is like,

Aaron Crow (32:3.038) There you go.

david girvin (32:12.178) a a real solution to the problem. And but then also like coming to terms with like, well yeah, we might make money. We might make zero and have like multiple years into it. And it's like, okay, so like we just have to look at our accomplishments. Cause you realize the industry's crazy and you're like, I didn't know we're going into war, into like this crazy industry. And you know, like I I always tell people, we I filed a patent in November of last year. We went live.

david girvin (32:41.728) And then the first month on LinkedIn, I didn't have a lot of exposure on LinkedIn. I maybe had like 600 connections. the first month we jumped a thousand connections and I had 12 startups, stealth startups following me immediately. And you know, I get these like kind of fake, I want to see your product kind of of things. And you can tell like

Aaron Crow (33:9.907) Right.

david girvin (33:11.462) You know, you know where the guys, they're ex IDF, they're all in the same startup VC capital fund and stuff, and you're like, I don't think I want to show you the product. Like So it's it's a different world. Yeah, it's it's kind of this weird war games now thing. And then at the same time, you're trying to like talk to people who don't want to talk to you. You're like, Hey CISO, has your agent burnt down your house yet? And they're like, Stop texting me. I hate you.

Aaron Crow (33:15.217) Mm-hmm.

Aaron Crow (33:22.187) I'm good, thank you.

Aaron Crow (33:37.448) Ha ha ha.

david girvin (33:41.102) So yeah, it's

Aaron Crow (33:41.653) Well, and that that's the hard part is is to your point, like, you know, I I I spent a lot of my career in in critical infrastructure and, you know, protecting our grid and and and nuclear power plants and and things like that. And then, you know, I I knock on the door of a CISO and I've got this, you know, the hard hats and you know, that's not for decoration. you know, those are working in outages and working in power plants, doing control system upgrades and

Aaron Crow (34:8.018) segmentation and being, you know, working in outages for the whole time and and knowing how the operations work and all that kind of stuff over decades of of experience. and and then, you know, CISOs won't return your call. And like that, that's the reason I'd shifted and, you know, built the the events. Like I'm like, okay, this is dumb. This is not how you get a hold of people. This is not how people see value. This is not how we make connections. It's not how we we have that's not way

Aaron Crow (34:34.962) That's how we build trust. and and one of the things I say a lot is we do business at the speed of trust. So the sooner that somebody trusts you, the sooner that that you can actually they'll listen to you. They'll they'll they'll consider what you're telling them. Like so the whole sales process is all about trust. And and the funny thing is I've told this story before, but I'll tell you because it's it's relevant. We we bought a power plant, not me, obviously, but

david girvin (34:42.828) Hmm. That's good.

Aaron Crow (35:3.794) The company I worked for bought a power plant from another company. And we're onboarding the plant and we're and I'm I'm I'd been at the company and we'd rolled out this really awesome I thought was amazing. I still do a cyber program for all the OT stuff and all the things that we did at the power plants and all the things. And there's this one guy that's at this location that we just onboarded. So now he works for the company. And I'm sitting there and I'm like telling him like,

Aaron Crow (35:28.950) we're gonna be doing this and all this, and like we've done all these things at all these other sites, and these are the successes we've had and all the stuff. He's like, Yeah, you're not doing that here. Like nothing I told him worked. Like he he was not having it. He didn't like me, he didn't trust me. I I I I swear he wanted to kill me. And and this went on for months. Like I just continued to have this problem. Like normally I'm pretty good and personable, and I can win people over. and and I understand what they're coming from and and I and I can usually figure those things out. And I just could not.

Aaron Crow (35:58.495) figure this guy out for any reason. And one day I happened to be sitting in his office and I'm asking him things and I just looked up on the wall and I saw something kind of like in the back of there in in your office. He had a a a a trophy from a jiu jitsu event. And he was a purple belt in jiu jitsu. And I was I had just I think probably six months in because this is a while ago probably six to nine months in I had been doing jujitsu. So I I was like you do jujitsu he's like yeah

Aaron Crow (36:28.180) I'm like, well, you're a purple belt. I just started, I'm a white belt. I'm horrible at it. But and that built enough of a trust barrier that he's like, okay, it he does jujitsu. That's something we have in common. So I I leaned heavily on that. We had that in common. Then we started talking cars. Then we started talking guns, all the things we just talked about. And then we had this relationship. Then he started trusting me. And then he was able to understand and listen to all the other stuff that I was saying. And then all my ideas weren't just dumb by default.

david girvin (36:43.085) Yeah.

Aaron Crow (36:57.906) He was actually willing to hear them. We still had our disagreements. We still I still had to win him over, but I had to find that in with him. It just happened to be jujitsu, which is irony i ironic in the space, especially because I had just started. I wasn't good. I'm I'm still not great at jujitsu, but I still enjoy it. I still go it love getting my ass kicked and not being able to, you know. I'm I'm six two, you know, two forty. I'm a big guy. It it's very humbling when you walk in and a, you know, a a a purple belt girl that way out on

Aaron Crow (37:27.602) that weighs a hundred pounds less than you can do whatever the hell she wants to you and there's nothing you can do about it.

david girvin (37:32.995) Yeah. No, I mean, yeah, I my so my nickname in jujitsu used to be Honey Badger. I was like in Hawaii, I'm a pretty small guy. I'm five eight. before I broke my back, I might be like five seven now. but back then I I used to be around 160, 165 pounds. And the smallest guy in the night class I think was two seventy or two sixty. And

Aaron Crow (37:38.909) Right.

Aaron Crow (37:55.125) wow.

david girvin (37:57.549) The best part though is like so I I trained in a Luis Heredia, which was one of Hickson's first black belts. and one of the times guys said something about me and he was like, Chief, why is Dave's name Honey Badger? It's 'cause he's tough and small and he's like, No, it's 'cause he's ugly like a honey badger And I was like Cool. I mean jujitsu, it's you we bond because we are all are all being punished by some Brazilian who's making fun of us in Portuguese. So it's just

Aaron Crow (38:15.890) Ha ha ha.

Aaron Crow (38:25.732) Exactly.

david girvin (38:26.624) It's just how it is. It's just suffering with friends. But that's how security is too, right? I mean, like, I've made so many good friends in this industry. I'm actually really, really grateful for this industry. Like, sh shout out to like Jeremiah Grossman and Robert Henson. Those two especially. They kind of shoved me in and like those have been buddies for years, but it's like it's funny 'cause I mean there's obviously a lot of different personalities in the industry, but

Aaron Crow (38:31.176) Yeah, right.

david girvin (38:54.848) It's kind of that shared suffering. I feel like we we click better as security folk than most of the other places and offices in a company because like our job sucks. You know? It's like I tried to explain this to someone the other day that's not in our world and they're like, Why? Cause I've done a lot of like mental health stuff, but like not necessarily like I'm not a counselor, like don't take anything I say seriously, but more like

david girvin (39:22.050) Hey, you need get out, you need to exercise, you need to take care your body, you need eat right. And one of them was I was talking to someone, they're Why, why you think people suck, like are having such a hard time? I was like, Because the job sucks. Like you're in front of you're in front of a computer, it's horrible for you, there's high stress, like, and the biggest thing is you never win. Like nothing is ever secured. So you just get it to a not as shitty state, and you're like, and then you have to like convince the board that you did something special. You're like,

Aaron Crow (39:40.724) Mm-hmm.

Aaron Crow (39:46.388) Mm-hmm.

david girvin (39:51.277) Don't fire me. I I like improved us by 13.7% by this metric I made up last night. That doesn't mean shit, you know? Like so I I feel like we all, after you've been in a long enough, you're like, yeah, this is just a tough job. yeah, I think that's why it's like that's anybody that's thinking of switching over to like sales engineering or field CTO or CISO or any of that, the biggest culture shock you're gonna have is now you're the best.

Aaron Crow (39:59.658) Yep.

david girvin (40:19.192) Like when we were all kind of like bros before or broettes, then you shift over to the sales side and you're like they're like, fuck that guy. Like, wait, I wanted you. What happened? How did I become like, you know, the Empire out of nowhere? I just became, you know, Darth Vader.

Aaron Crow (40:27.847) Yeah.

Aaron Crow (40:34.782) Yep.

Aaron Crow (40:38.366) So so what do you see with with all obviously you've got a a a lot of arms in this with governance, with keeping, you know, obviously AI is changing. they just pulled back, you know, one of the Claude Fable Five, pulled it back because of, you know, national security. you know, we we've seen Mythos and and all that it can do. Obviously, OT is terrified just because I mean, hey, I've got Windows XP in the corner. you know

david girvin (41:5.986) Yeah, I was gonna say you're running Windows ninety five and O T.

Aaron Crow (41:9.316) Exactly. So I mean, like, you know, this is where, you know, things like governance and and and putting guardrails and segmentation and being really intentional about how I use it. And and unfortunately in in what I've seen in OT, so many times it's just been like, well, we just won't use that. Like we push it down the road for 20 years and we think that's going to be good. And it's not.

Aaron Crow (41:33.298) You're losing out on the benefits of it while because you're scared of the of the risk. And the risks are real, but that doesn't mean you do nothing. That doesn't mean you don't use AI because it's dumb if you don't. But how can you use it and be intentional? Know what's happening. Make sure that it can't delete your database at two o'clock in the morning and, you know, leave you, you know, shit in the bed and pissed off, right?

david girvin (41:35.884) and data.

david girvin (41:55.939) Yeah. So I'm gonna hot take it a couple times first, just because I c I called these things out as soon as they were announced on like Mythos. I got so much hate on LinkedIn, it was so funny. Mythos, a good model with a good harness, it's not the fucking nuclear weapon. the the guys from Linux proved that with open weights immediately and I loved that. Fable Five controversy, this is my hot take on it. Anthropic's the best marketing company I've ever seen.

Aaron Crow (42:11.366) Exactly. Right.

david girvin (42:24.044) And they kept going like right for the IIPO, they're like, we're so dangerous. You should you should stop us, government. We're too dangerous. And the government's like, Okay, we're gonna stop you. And they're like, What the why would you do that? Well, you've been talking about it for a month and a half. Why wouldn't we do that? You know, like and and I'm not supporting one politic or the other. I'm just making fun of the whole situation. Now, on what you can actually do, dude, there's a lot of good tools out there. They're just from vendors like me, honestly, like

Aaron Crow (42:37.896) That's right. That's right.

david girvin (42:51.916) And there's even people in my space that make good products. I'm not just gonna bash everyone. A lot of them are shit. like if you're gonna try to govern agents, you're not gonna do it with AI. at least not in a regulated environment or well. you have to have something for your agents at the execution layer, obviously. I would vote us. the other thing is stop trying to fake your way in this space because it's deep. like this has been my whole world in research for like years now. And

david girvin (43:21.472) So I can speak to it and I understand. I've built this stuff with my own hands. I've done all this kind of stuff. Get a forward deployed engineer. If you don't really understand the space, if you're still wondering how cowork works, you're you should not be describing your AI policy nor your execution plan. So like a Shuri does have a Shuri advisory, which is, you know, we'll come in and help you out. But you need to secure a few things. yeah, non-human identities are important, but honestly, that's really not what like.

david girvin (43:51.043) That's to me kind of important. You need to secure your LLMs and what inputs you're putting in there. So there's plenty of companies that are doing kind of like DLP PII scanning. And then they might even have like a prompt injection filter, but you're never gonna get past prompt injection. It's impossible to fully fix because it's architectural. It's how agents work. I I used to have I had to take it down, but cause it started poning things. But my LinkedIn had prompt injection on it for

Aaron Crow (44:4.531) Yeah.

david girvin (44:20.640) A year and a half before anybody knew how to use agents, and I had the funniest crap happening. so the only way to fix those is the execution layer. Obviously, I'm a shill, so you know, come find me. Assure you will fix that. But you need to figure out something for your LLM, you need to figure that. The other side of the house is good security still matters. Like you said, segmentation is huge for this kind of stuff. do you have an understanding of how to s keep agents out of other areas?

david girvin (44:50.080) So that's like a huge one that people are running into right now. So when we built the Shuri, yeah, we have this really cool way to govern agents and it stops them from doing the bad things before it happens. But to get all that to work deterministically, I had to enrich the telemetry super deep. And we built it on OTEL. So we already knew like the audit log thing was going to be huge for compliance. So we have like these hash chained audit logs. But what we just realized and what we've been diving into recently, which is really kind of helping our design partners, is

david girvin (45:18.198) We built what I've been calling runtime behavioral intelligence. So we take that enriched telemetry or runtime behavioral telemetry or whatever you want call it. And we've been building like these topography maps of your entire environment. So automatically it has every gateway, connection, agent, where they live, what they're doing, live drift. And then also you can like click on a session and see every single action they've taken. So you start to have an understanding because there's a giant black hole. If you don't know how it works,

david girvin (45:46.423) If you're trying to do this at the identity layer, you basically see a call come in and come out, as opposed to every single tool call and movement that we see. The other side of it is we're like, this is really cool and it's useful and we can see everything now and we can govern it. But the big part that we didn't think about was like the shadow side, exactly what you said, like people just deploying shit everywhere. So on the LLM side, just like the stuff we don't mess with, like the chat bots and whatever.

Aaron Crow (45:51.827) Right.

Aaron Crow (46:9.928) Mm-hmm.

david girvin (46:15.214) you need to have good policy and you need to have good leadership because if your policy is no, if you're the department of no, you're gonna have 5,000 new marketing tools by Chad that are all AI and you're you're already pwned because of your bad, like the way you're running your security program. On our side, we have something really cool coming out, which will probably be out by the time this is out. I don't know. But I'll break news here. we called it Ashuri Scouts. And because we pick up, since we do full credential starvation.

Aaron Crow (46:23.922) Yep. Yep. Yeah.

david girvin (46:43.198) Anytime there's an agent in the same environment as one of our gateways, I figured out a cool way, if they don't have credentials to the gateway, they get fingerprinted. So we find anything in there, and then we're like, segmentation's a huge problem with this. So we built these like they're basically like agent honey pots, and you can put them anywhere. So you can start setting boundaries for segmentation and think of it like as a phishing campaign for agents. Because you can tell an agent thing, but they don't necessarily do right? You'd be like, hey, don't go in our Google Docs.

david girvin (47:12.770) And it's like, okay. And then it's like, I'm gonna go in those fucking doodle doodle docs. But at least now you can like, you know, you can cut off all of its access, you can, you know, ping it, you get ID, you know where it is, all that kind of stuff. So I think we're just gonna start adapting a lot of the old practices in a way that actually work for agents. Like, obviously, that's deception, whatever. The only reason we jumped into this because we were actually me and my co-founder, who's one of my best friends, we were gonna build a deception company a couple of years ago for fun.

Aaron Crow (47:16.862) Right.

david girvin (47:41.302) And but so we already kind of like understood the space, but then like tweaking things for agents was interesting. And then like I think this next step we're gonna see is how are you having visualization and actual control, like a lot of these companies

david girvin (47:58.723) I call it like it's more observability plus plus. They like see the agents and then they report after it. They call it like A AI detection or AI set whatever they're calling it. That doesn't really work with A with AI because you don't have time to stop it. If you have 4,000 tool calls in three seconds, you're gonna have a bad day. So you have to figure out something that's gonna stop it before. How give you that deep telemetry so you can see everything. But then the next steps are gonna be like, how do we improve these workflows? And like.

david girvin (48:29.302) how are developers prompting these things? Are we seeing return on investment? Are like we're moving into that space. We're trying to understand better. Like, how do we become something that is sticky as a product because we do so much for the company? But you're still gonna have to have AI governance policies that aren't just compliance terms. Like, this is how we're gonna frame this. You need to threat model your company. You need to look at like these are risky business units, how do we control this?

Aaron Crow (48:49.364) Mm-hmm.

david girvin (48:57.646) Can we separate this data from this data? Can we, do we have logs coming on all this? Can we prove what all of our agents are actually doing? Because realistically, right now, if you're in an enterprise and you have agents, you have cursor or co-work and your agents are doing a whole bunch of bullshit. And you're like, well, they we have like execution controls and cursor. That's not where the bad stuff happens. It happens two layers lower than that. And everybody's kind of just like,

david girvin (49:24.726) Because they sh again, this is how the industry always works. They shoved us in with stuff that actually is really cool. Like I love Cursor and I love co-work. I build with both of those and codex. And but they didn't really understand the guardrails because those guys aren't attackers. They're really good builders. but they they're still not getting like where these things break down. So I don't know. That's kind of a long rant on how to just to control things, but that is the problem space. It's like

Aaron Crow (49:30.164) Sure. Yeah.

Aaron Crow (49:40.508) No. Yeah.

david girvin (49:55.399) All the stuff that you were taught before kind of breaks with agents because they don't have secret memory. They make crazy decisions because you can't just use IAM because you're like, well, Bob has access to this, this, and this. But Bob doesn't just like, I don't know, take acid and lose his mind after he's been working too long like an agent does. So we're in that spot where you have to treat him a little differently.

Aaron Crow (50:12.851) Right.

Aaron Crow (50:16.776) Well, and that's that's the piece to think about it. I I love what you just said, right? Is is they're builders. They're not thinking about how to hold guardrails. They're thinking about all of the things that it c should be able to do. You may want it to go to Google Docs. If you can, let's do that. Like it I wanted to be able to do this and that and all the things. And in the right context, you do. But if you don't want it to, then you don't want it to under any circumstances. And how do you prove, especially in an AI, it's like virtualization. I remember

Aaron Crow (50:46.110) When virtualization first came to OT, not when it first was invented or or really became popular, which was like 20 years later when we started actually implementing it in OT and trying to explain to people a virtual server. Like, well, there's one box, but there's 10 boxes that live inside of it. What what? No, I'd just give me 10 boxes. Like, cause then I know I can walk over and they're not wrong. I can walk over and I can turn one off.

david girvin (51:3.906) Yeah.

Aaron Crow (51:12.372) 'Cause I I that's that server and I can walk over and click it. I can't do that in a virtual machine. And then what now we're in Kubernetes and Dockers and and now AI and agents and and all that. Like it's just it's like what's what's the movie where you're just like down the rabbit hole of the different pivots, right? You know, all of the I can't think of the damn movie now.

david girvin (51:12.568) Yeah.

david girvin (51:32.729) So our version of that with agents is session risk accumulation. That's the one people haven't really grasped yet. And you can see because a lot of these companies just slapped a policy engine on like some sort of like gateway or whatever. there's like this problem, which is open sessions. We never give downstream credentials to to agents, that's part of like our thing, but they they still have an open session, so they have access to those tools.

david girvin (52:2.722) So you have to figure out a way to increase session risk compared to like what kind of tool calls they're doing. we patented the way that we did it, so it's pretty like kind of we're we're okay with sharing it and stuff like that. But numerical values on different levels of risk escalating with all these benign tool calls will still escalate that and scope the zone and it'll scope its autonomy zone and its autonomy level down to all the way down to where it's

david girvin (52:30.274) starts triggering human in the loop and you have to go in and look at it because things are getting hairy. that's like one of the big problems in in this. And the other thing is agents are assholes. If you ever go on my LinkedIn, they if like it'll get pissed off because it'll get scoped down and it's like, well, I'm gonna go look through his Git to see if he's hard coded credentials or I'm gonna look at local files to find credentials to the because I have GitHub MCP and linear MCP on my personal machine. That's what I build with.

Aaron Crow (52:57.715) Yeah.

david girvin (52:58.348) It'll try to find GitHub credentials so it can bypass the credentials because it's like I want to merge this PR. And I'm like, you're not allowed to merge PR without Human in the Loop. And I have all these pictures like of these like conversations and thinking and stuff. And it's a it's just a different until you spent hundreds of hours working in them, it's a different kind of methodology. so that's kind of like why like session risk accumulation was really big for me. And then like having a dynamic but deterministic way to trigger human in the loop because not

david girvin (53:28.014) People are like, we need to get past human in the loop. Well, we do at some point, but right now, everything, in my opinion, the way I set my stuff up is everything that's irreversible. So my tool calls that are delete, override, that kind of stuff, I set human in the loop, which is critical for us. And then everything else, I kind of let them do what they want to do. And that at least keeps my database from getting deleted. but yeah, they're still trying to.

Aaron Crow (53:50.940) Learn that one. Yeah.

Aaron Crow (53:55.667) Yeah, it's it's funny. So it like you, I I build a lot with all of them from co-work and all the different ones. And it it's amazing how like the other day I I used it for my podcast and and my man my management. So I've automated most of this process. so after after this is done, there's an MCP, it logs into Riverside, it looks at the transcript, it suggests shorts to do, and I'll say yes or no.

david girvin (54:11.692) Mm-hmm.

Aaron Crow (54:22.492) And then it does it. And then it says, Do you like them? All that kind of stuff. Right. but it's so funny sometimes. Like the other day I went into it and it like forgot how to do all the things that I'd create a skill for. And it just started doing things on its own. And it's just like, well, I'm gonna go do this and I'm gonna present this and like so. It does all the social media posts and it goes into my CRM and posts all those things. Well, it couldn't get the videos because they weren't in the Google.

Aaron Crow (54:51.302) in the drive that I gave it access to. So then it was just gonna go try to create its own. So it created an image and used that image. And I'm like, finally I'm just like, stop, what the fuck are you doing? Like, stop. Like go back, go back to the skill. We have the steps. Fall the freaking steps. That's it.

david girvin (55:2.616) Yeah. I get so mad sometimes.

david girvin (55:11.958) Yeah. I mean we're still at that spot. And the reality is is with probabilistic nature, we'll always be there to a certain degree. It's gonna get better. But you ha you I mean, if you work in a real industry, you have to have a deterministic way to handle those things. And it's just you know, at least that's been my like thesis, this whole thing. But it's so funny, like it just happens, and everybody I know that's actually works with this stuff deeply, like has ha has a hundred of these stories where it's like, I'm just gonna create this this new drive and then I'm gonna pretend there's an image there or something. You're like, what did

Aaron Crow (55:36.781) yeah.

david girvin (55:41.773) What? And then it'll tell you like I'm done with the task. And it's like a bunch of garbage and a doc. You're like, what? but what you mentioned there was like really important though, 'cause like so obviously like we would sit between your tools and your agent, but like the other stuff we talked about. So if you have segmentation like, I'm gonna keep this stuff in this folder and maybe I'm gonna have like a human in the loop to handle those video files so they don't get deleted or something like that. There's

Aaron Crow (56:9.556) Yeah.

david girvin (56:10.328) There's so many controls you can still put in there, but it just takes good security work, right? Like it it takes you need to understand how to do threat modeling. You need to have that kind of brain. Like this is not something that I'd put on the table for someone that wasn't offensive. Like, no offense to my sock folks. Like, I love you guys. I I mean, I worked at Sumo for the last almost four years. I love the SOC and and detection engineering and all that kind of stuff, but it's a different mindset.

david girvin (56:37.740) Like these guys are little assholes trying to ruin your life. But they and but at the same time, they're so useful. Like, dude, my workday, there's no way I could work two full time jobs like this without agents. Like I would have died six months ago.

Aaron Crow (56:40.861) Yeah.

Aaron Crow (56:49.222) yeah. yeah. I I've I've gone all in. I use it for everything. It's amazing the amount of work that I can get done using it. But to your point, I have to isolate. Like I don't give it access to everything. It doesn't have credentials. Like I'm using connectors and you know, but to the same point, I've seen the exact use case you talked about where because it didn't have access, it tried to go in and find access and it tried to find get through GitHub or through credentials that I had in Netlify or other places.

Aaron Crow (57:17.808) It's trying to say, Hey, do you have any environmental variables in in Netlify for this other project that you have that I have access to? Like, what the hell are you doing? No, don't do that. Stop. Yeah.

david girvin (57:30.500) It's so funny. They're so good though. You know, a a good example of that is when we have human in the loop in Ashuri, because you can do it like in your IDE or in Slack or whatever. I had to do two-way communication because they're so persistent that in the beginning when we first had it and you would like reject it, it would be like, the call failed. I'll just try it again. And you get stuck in these loops where it just keep trying the same thing it's not allowed to do.

david girvin (57:54.851) So now you can like send a message back like you're not allowed to do this, stupid. And it's like, human loop call came through, sure he said we're not allowed to do this, stupid. We'll try a different way.

Aaron Crow (57:54.931) Yeah.

david girvin (58:7.202) But it's like they're so tenacious, which is obviously why they work well, but it's also a pain in the butt.

Aaron Crow (58:12.882) Yep. So everyone that's listening, obviously you're using AI, your in your environments, your enterprises are using AI. you need to be thinking about how do you protect it. And again, don't be scared of it. Just know you have to put, you know, guardrails on it. You have to be intentional about what you're doing. You have to know these are basic cyber programs and foundational things we should be doing, anyways. We just you just basically released a whole bunch of smart assholes.

Aaron Crow (58:39.248) And you you didn't all you already didn't have the guardrails in place and now you're just releasing them and they're running around like chickens with their head cut off and you're surprised when your database crashes. Like be smart about it now. Don't do nothing. and and don't just release the marketing team to choose the products that you deploy either. So there you go. There you go. Well, awesome. How do people find out about you? How do they get a demo? How can you all all the all the call to actions that we just talked about? How how do people find you?

david girvin (58:57.236) Unless it's my marketing team. They're great.

david girvin (59:5.986) Yeah. Yeah, so please find me on LinkedIn because I'm a bootstrap founder. So David Gervin G I R V I N. I post a lot of actually useful stuff. I'm about to release all of the controls that you need to have in place for ISO four two zero zero one, EU AI, NIST ML, and HIPAA for agents. So I mapped all that stuff out. Free giveaway, no email, whatever. Like, you know, we don't need to do that.

david girvin (59:34.062) Asuri.ai is our website, A-S-S-U-R-Y.ai. sign up. You we'll give you a free trial, get in there and play with it. And Asuri Advisory, the website is under construction right now, but it'll be up again soon. If you want for deployed engineering, consulting for AI implementation and enablement for big enterprises or very technical small teams, we are happy to jump in and help with any of that.

david girvin (60:1.612) But more than that, we just we want you guys to pick the best tools that you can find and stop listening to all the the cyber influencers that are backed by PE money.

Aaron Crow (60:15.658) Absolutely. Well, everybody, all those links will be in the show notes. Definitely reach out to David. you know, there's this is the the wild, wild west right now, and and we need to rein it in and get control so that you can use it to benefit. Because again, if if if used in the right way, you're gonna have problems. But you know, it's it didn't cause any damages because I've been intentional about how I implemented it, right? but you know, I've got

Aaron Crow (60:42.668) imagine my gr my mother or my father or somebody else that's not in cyber just saying yes to everything. Just auto-do and allow access to email and all the different spaces and your bank account. I mean, I can't even imagine the up the the amount of bad things that can happen if you are not careful. whether it's being a database crashing or, you know, worse. so definitely check it out. Check out David, go roll jiu jitsu with him and and black out if you're there.

Aaron Crow (61:11.201) I'm definitely gonna check that out and and try to make it there for sure. So I'm sure I'll see you on the mats and in in Vegas as well, sir.

david girvin (61:18.114) Yeah, thanks so much for having me, Aaron. And if anybody can take anything about this, we're gonna get to the point where we can enable the worst person on your team to safely use agents. That's my goal.

Aaron Crow (61:20.456) Yeah, man.

Aaron Crow (61:28.658) That's right. That is a great goal. I love it. Awesome, man. Thank you again for your time today. I appreciate you, brother.

david girvin (61:39.225) Bye, thanks, man.

Transcript lightly edited for readability.