Latest / CIO Unscripted / Dean Sapp on AI's Speed Advantage in Cybersecurity and Legal Tech
Transcript
- Dean Sapp: AI is very, very powerful. But we what we also learn from AI is it's often unpredictable. And if you give it a mission, it often will disregard future commands to to stop the mission. It's going to do the mission and it thinks that any any other communications between stop and start. or start and stop is is trying to to get it off course. So what we have learned is AI is incredibly obedient to the first command you give it. Niel: Welcome to CIO Unscripted. Nice to have you here, Dean. I'm my guest today is Dean Sapp. I'll have him give his exact title, but when it comes to cybersecurity and its intersection with AI, I think Dean Sapp is the smartest person I know, and I know a lot of people. So that's saying a lot. Anyway, I've known Dean for years. He's had a fascinating career. And I think what he's doing is on the forefront. with AI and security both, long track record in cybersecurity. Dean, I appreciate your being here. I really do. I know this is quite a bit of of a sacrifice for you as busy as you are, so I appreciate your being here. Dean Sapp: No worries, Niel Thank you for the invitation. It's a pleasure to be here and just talk shop and you know, this AI journey, security journey, kind of can it's joining together with how do we even just sleep at night? Niel: All right. With that, Dean, why don't you just come a g give everyone a brief overview of your career and how you've sort of ended up being where you are at this point in your career? Dean Sapp: Sure. Well, I would say my parents probably classified me as a mischievous child. You know, I I may or may not have had a a challenge with keeping the rules and you know doing things that creative people like to do, like taking things apart and figuring out how they work. So I've always been very interested in computers, electronics. I built several computers as a as a kid and as I got older and my fortunately my parents had early Apple computers and I was just super interested in in how they worked and using them. And of course my dad was he was a very, very creative man. He he's brilliant. and he has several patents in his name. And so I ultimately wanted to do what he did. I wanted to he ran several boat companies and was an inventor and did a lot of really fascinating entrepreneurial things. and so as I went through my high school days, I had an opportunity to take AutoCAD classes and technology classes. This is before STEM was a thing, but I was very fortunate my high school had computer classes and things. And so I learned networking, early Novelle days, and I learned all of the DOS programs and early Windows variants and so I was able to get a lot of hands-on expo exposure and and frankly I I love drawing things. So I was doing mechanical drawings for AutoCAD, like engine parts and and things that were were interesting to me. I my favorite magazines were popular science and popular mechanics and so I would just go check those out from the library or or whenever they came to our house I was always interested in reading those. But anyway, so my my high school had these AutoCAD programs, and I had a chance to do computer aided drafting and competed in Florida in in computer aided drafting, aeronautical engineering, and a a couple of different programs and ended up placing second and third in the state of Florida in these different events. And I'm like, this is really fun to me. I love engineering and building things. So I just parlayed that into into my career. I worked at Novell with the the experience I I gained and my first job after I went on a a church service mission for my the Church of Jesus Christ Latter day Saints and spent two years in inner cities of Houston, Texas. But once I was done with that service and again loved it loved the people of Houston and great way to to serve and get to know your neighbors I came back and and went to college and was I had a academic scholarship to a university in Central Florida and it lapsed while I was out serving my mission. But then my grades were good enough when I came back. I I got a full ride again to go to the community college where I was. it's now a full on four year university now, but was able to get you know associate's degree. And then transferred out to Brigham Young University. And like any anyone with a crazy brain, I had a real hard time going to school all the time and not working. And and so I ended up working full time almost my entire career while going to school part time or full time. So I was, you know, double dipping, but I always like a challenge and my brain likes to be in hyperactive mode or or asleep. So those are my two choices. Niel: Very binary, huh, Dean? Dean Sapp: You know, full s that's right. Full speed ahead or resting. So anyway, just was has been able to to leverage the passion for learning and and breaking into stuff. You know, I may or may not have gotten in trouble as a kid many times for for breaking into things that I probably wasn't supposed to. but you know, I also like to learn the boundaries and how boundaries work. So anyway, so I spent a couple years at Novell. And Novell had a tremendous program. Anyone who was an employee, contractor, or full time, they would let you check out any of the coursework for their classes and do it take the tests for free. And so I like free. I think everybody likes free. And so I while I was working full time and going to school part-time at BYU, I took I think it was eight or nine of the Novell certifications. So learning how to run a Novell directory, border manager, NDS for NT, Zenworks, all of these different tools. But you had to understand the security principles behind those tools and how to do things safely and securely and using all kinds of token ring and all the good token read jokes, who's got the token, you know. so during that time I I was able to I was really fortunate to get nine certs I was a master certified Novell engineer. Well, and with that experience, come to find out there was a very large law firm in Florida that needed a Novell expert. And I'm like, Well, I would I'd been at BYU for for a while and and frankly was getting bored. So I I applied and or actually had a recruiter just reach out to me and and this law firm was willing to fly my family back to Florida, which Here's where from. And so I spent the next eight years at this law firm. r well, the first six months was pretty purposeful. They had had an engineer that had struggled to keep the network functioning properly, and he just walked out. So he had spent like a a full three-day weekend. The directory was down, mostly because he he followed some bad instructions. And so no one could log in for a couple of days. So my first job was to get it so everybody could log in again and and that was fixed immediately. It was simple to just build a new replica of the directory. but then building all of the the scripting and the things to automate so that the network would just run on its own. This was like before the days of of updates, I just wrote a bunch of cron jobs that allowed me to script. Niel: Sure. Dean Sapp: all the maintenance you needed to do on the directory service to keep keep it clean and to make sure that there weren't you know errors or or objects that were in a you know shouldn't have been in the directory. So anyway, so it literally took just six I think six months to get everything just humming for a very large law firm. And then I'm like, what else can I do? This is boring. So you can see there's a trend here. I can't I've got the attention of a squirrel, Niel: Yeah. Dean Sapp: attention span. So I ended up just saying, What else do they what else do you got? What what can I do? Well, at the time, a very large, like a top ten bank was one of the the law firm's largest clients. And they said, We have these security questionnaires that we keep getting and we need you to build a security program to satisfy all of these. I'm like, that sounds fun. Let's get to work. Niel: so Dean, let me interrupt you. About what year was this? Dean Sapp: you're not allowed to ask questions like that. But Niel: sorry, sorry. But Dean Sapp: over twenty years ago, how about that? Niel: Interesting. So the banks, financial services were already doing security questionnaires twenty years ago. Interesting. Dean Sapp: They were and they were asking for things like PGP encrypted email and for portals and whole disk encrypted Blackberry devices. So we were everybody had every attorney had a Blackberry, the data had to be encrypted, they had to be able to send encrypted email over that client over that that BlackBerry client. So building all of the firewalls, all of the gateways, all of the The exchange hardening for the so basically I I was told, hey, build a program. I'm like, well, you know, what's my budget? You don't have any budget. Just go make it happen. You know, you lawyers like to see things happen just, you know, out of thin air and creativity. So I I learned a lot about scripting and automating tasks and configuring PGP universal servers, create hardening border manager servers. When I first started, every computer had a publicly routable IP address. And I'm like, my goodness, what are you guys doing here? so of course we had all kinds of problems on the network and so dropped in reverse proxies and caching and and 'cause frankly, lawyers do need to do a lot of research and they need Niel: Yeah. Dean Sapp: to get to some sites that may not always be ideal, but if you're defending a client, you know, you have to have that. So I Niel: Yeah, Dean Sapp: I my Niel: right. Dean Sapp: eyes were somewhat open to the realities of of the world while also trying to defend it and trying to do the right thing. and I also learned a very important lesson. If someone at the firm pushes back extraordinarily hard against security controls, you know that there's a problem. And and of Niel: Okay, got it. Dean Sapp: course, I was dealing with with a particular individual who was just always giving me grief about just basic security practices. Well, the FBI came in, took them away about a year later. So you you never know what's what's going on. But the one indicator if somebody seems to be completely opposed to anything you're doing security wise, there's probably a reason. So anyway, I I spent eight years building that security program and and helping to meet these bank level obligations for encryption and protection and role based access control, all of those things. And and then my wife at the time wa happened to get pretty sick. And I had actually applied for a job and had forgotten completely about it. And then I I was flying back to visit my mission reunion. So all the missionaries that served with me in Houston, Texas, They all we had a ten year anniversary and so I had plans to fly to Salt Lake to go to my mission reunion and then I had tickets to General Conference, which is a twice a year conference for members of the of that faith, to get together and and listen to to doctrine and teachings. So I had plans to go to both of these and so while I was there, they're like, Hey, you wanna interview for a job? We'll fly ya out. I'm like, Well, I'll be there tomorrow. So why don't we just catch up? So we ended up just interviewing all day from at the that large organization and they offered me the job. And so when I got home to Florida, I found out my my wife was really sick, didn't understand how sick she was, and she ended up needing to have emergency surgery. So all of that being said, it it kind of worked out perfectly. We listed our house, sold it for cash the next day. Just a whole bunch of things happen that I've no explanation for. and so, yeah. Niel: So so Dean, what job was it that you'd interviewed for? Dean Sapp: So I had interviewed for essentially a virtual CISO position for Niel: Interesting. Dean Sapp: a very large not-for-profit. Niel: Okay. Dean Sapp: And and I actually had never even planned to attend to enroll or not enroll to apply, but I was actually, oddly enough, I'm you can probably tell I'm a pretty religious person. So I was actually at church and the bishop of my church pulled me aside and said. this company wants you wants applicants for cybersecurity. You're a cybersecurity guy. You should apply for it. And I'm like, I'm pretty happy where I am, but okay. So I he promised he made me promise I would apply for the job. I did. I didn't hear anything for a year, and then finally, you know, these these different events all aligned, and so I applied, interviewed, and they offered me the job. Niel: Yeah. Dean Sapp: And the really ironic part is that same trip, my the CTO at the law firm also told me he was leaving and going to a law firm in California. And he asked me if he ought if I wanted to be his his security director and IT director. So I literally flew to to California first, interviewed, got Niel: Okay. Dean Sapp: offered the job, flew the f to Salt Lake. Interviewed, got offered the job, and went back home to Florida to find out that my my wife was super sick. And so I'm like, Well, she was from the Utah area. So I'm like, we're going to Salt Lake City. Niel: Good for you. Dean Sapp: Yeah. So ended up building a cyber program over the next eight years, hiring, you know, we hired dozens and dozens of risk assessors, soc, and we built our own security operations center. We built Basically a full cybersecurity consulting company for these legal entities of this Niel: Got it. Okay. Dean Sapp: large not for profit. So just Niel: Okay. Okay, that makes sense. Dean Sapp: yeah, so you know, eight years of the law firm, roughly, eight years at a large not for profit building a full MSSP so that any of the the business entities of this large corporation could then just subscribe to You know, manage sock, instant response, red teaming, you you name it. So the Niel: So you you built a shared security service organization for a diverse enterprise. Lots of different Dean Sapp: That's right. That's right. A big internal multinational Niel: businesses, types of things they were doing. Fascinating. Dean Sapp: Absolutely. Niel: And so to go back, your career track's been in cyber, right? Dean Sapp: Pretty much my whole career. Niel: Yeah, but I mean, but it that happened because somebody needed to answer a bank questionnaire, right? Dean Sapp: Right. Well and even before that and Novell, someone Niel: It's just Yeah. Sure. Dean Sapp: needed somebody who was willing to try and break into the RSA encryption Niel: Okay. Dean Sapp: and try to break the client apps and and make them do things that they weren't supposed to do. So it's just, you know, being told, Hey, you have free reign, break this stuff like that sounds fun. Niel: It's just it's it's fascinating to me the the people I know like you and the podcasts I've done, the interesting paths we've all taken to get where we are. It's like it's like these sort of these Dean Sapp: And a lot of times just timing. Niel: opportunities and I people ask, Hey, I I need your career advice, Neil. I say, when you see something, go for it. Yeah. Yeah. Anyway, sorry for the interruption. Dean Sapp: Go for it. Exactly. Exactly. I would just add that no, it's good. And and I would add, you know, I'm a pretty religious guy and pretty open about that. You know, anytime I was considering a new job opportunity, I would think about it. I would do like a pros and cons spreadsheet. Niel: Yep. Dean Sapp: And then ultimately I would make a decision and I would pray about it because I wanted some some additional guidance from someone Niel: Yeah. Dean Sapp: smarter than me. Niel: Yeah. Dean Sapp: And of course I always usually, you know, talk to my parents because they're super smart people. But anyway. So yeah, just you know, you're right, diving into opportunities. And and you wouldn't it wouldn't surprise me, or maybe it would surprise me or surprise other people, how often in my career things just happened in the right sequence. Like just before I applied for the job. I had asked, I had gotten interested in digital forensics. So while I was with Ackerman, I'm like, we had a we had an e-discovery team and e-discovery practice. Ralph Losi was a very well re recognized e-discovery attorney. And so we were building a practice and we had some very large government clients that hired the firm to deal with like Ponzi schemes. to do the evidence collection, to do the legal paperwork for the external legal work that the government needed. They of course have all their own attorneys, but they needed some of this to be done by external counsel. So I had said, hey, I really want to go learn about digital forensics. I want to understand how to get to free space and slacks and deleted information on a hard drive and all of both network-based and hardware-based forensics. And it was that very class that gave me a whole exposure to ciphers, encryption, all of the salt and hashing and all the differences between those different technologies. And come to find out, one of the biggest sections on the interview had to do with understanding digital forensics, which at this time was a pretty new field. Niel: Hmm. Dean Sapp: And so even though it it actually was I didn't do anything with that experience after I got the job because they had me focus on other legal areas and other cyber areas, it was the data I needed at the time to become the top candidate. And so that's been my career experience is that if something piques your interest, double click on it, dive into it, Niel: Yeah. Yep Dean Sapp: and and and do take a lot of the Know the motivation needs to come from within to learn new things. Don't wait for someone to ask you to figure something out. Go figure it out because it's cool to figure it out. one of my favorite experiences from pen testing was I I took a class on web application pen testing while I worked for this big multinational. And the day I got back from cla from the the class, I went into the office for like 30 minutes. At the end of the day, and a senior developer pulls me aside and he says, I have this new web app that I've been trying to talk with you about all week. I know you were out. I have a new web app that I want you to test to make sure that it's safe so I can push it into production. And so Kevin Johnson was the instructor at SANS that was teaching 542, which is you know advanced web app pen testing. And and Kevin said one of the the most important things you can remember in like a dot net app is that input sanitization and output encoding are often problems and the single tick, the single tick character, so the far top left of your keyboard, the single tick character Niel: interesting. Dean Sapp: is one of the most powerful that you can use for a pen test. So I I asked the developer, so what was this coded in? it's it's a.NET Stack app. I'm like, cool. put let's put a a single tick at the end of the last parameter field. I dropped it in, I hit enter, and it it it aired out and dropped the admin super user name and password to the screen. And I'm like is is that supposed to happen? Were you were you expecting that to to show me the username and password for the for the system? He's like, no. All right, you might want to sanitize your inputs and your outputs. But the funny thing about it is that one character happened to come because I had just gone to a class, I had just paid attention, somebody pulled me aside and said, Can you look at this? I'm like, Sure, I got five minutes. Come to find out that was the shortest pen test I've ever been a part of to get root access to the screen. But fortunately or unfortunately, it gave me a reputation at the company that you you better get your stuff in order so you're not embarrassed in 25 seconds by by the security expert. Niel: Yeah, because Dean Sapp: And it was just complete happenstance. Niel: Sure. Dean Sapp: Then I went to the class, I remembered what Kevin said. It happened to be the app type and the stack. And I and I sort of got like GOAT status in like three days. but then it actually happened again a couple probably a year later. They they were like, we need to know if our databases are secure. And so I'm like, well, great question. I don't know. And so I I went online and found a bunch of research by Chris Gates. who had the Metasploit mix-ins. And then I did a bunch of research by what is it, Alexander Cornbrusk, who did red dot security, Pete Finnegan, who wrote the Oracle database hackers handbook. So I just did a bunch of research on on the most current methods to crack into ninei or other Oracle databases. So I was just doing all of this research and I'm like, well I better get permission. So I asked the portfolio director, can I have permission to to test this? And he's like, sure. Yeah, just don't don't take it down. And so I did a bunch of research, pulled down the Metasploit tool. and I I was have actually having a problem with Metasploit running Chris's Oracle Mixings tool set, which is basically it's it discovers the the Oracle listener and it It does a bunch of things to help you find vulnerabilities in the system. Well, it wasn't running on my instance of Linux. So I I reached out to Chris on I'm like, hey Chris, I'm kind of new at this. Wanna could you help me out? I I'm getting this is the error code I'm seeing. And he's like, yeah, you need to go update some a couple of libraries. I didn't add that to my documentation, but here you go. He literally great support from a from a a black teamer, you know, or he's probably a gray, Niel: Yeah. Dean Sapp: a gray hat because he was on the good side, but writing new exploit tools for Metasploit. So literally I got it working, and then I got the signed pen testing approval doc. And then within six hours, I had the Oracle database credentials. Cause I'd be I was able to exploit the listener. get into the or database and then basically see all of the things that I I wanted to see. And it was just from curiosity, asking for help, and then just ha being stubborn about it. I'm I'm pretty stubborn, you know. I like there there are I can I can push myself. So anyway, so that's a really long answer to how I got about halfway through my career. Niel: All right. So you've been a vCISO , for this portfolio comp enterprise. So what happens next? Dean Sapp: Well, I you know, anytime you're in a large enterprise, big big company, I think you find that things kinda move slow. And as you can tell, I'm not a slow moving Niel: Right. Dean Sapp: person. I I get bored quickly. I ask for challenges. Once those challenges are done, I I move. What else? What's next? So I had an opportunity, again, one of those situations where I have a I have philosophy. Every day is a job interview. Every day you never know who you're gonna interact with or someone's gonna call and you're gonna you're gonna they're gonna ask for an interview. And I typically would take them because number one, that's really good practice. Number two, you you learn a lot about what you don't know by talking to other people who need expertise. So you know, someone reached out to me for To do a they wanted to do a cybersecurity startup for lawyers. And I'd already spent eight years building a cyber program for Ackerman and eight years for this big multinational. And one of my portfolios at the multinational was the the finance and legal portfolio. So I was working with attorneys around the world and trying to secure their environments. So I'm like the idea of building a cyber startup. just to protect lawyers sounded really fun to me. And so a company called Brain Trace was created and they brought me on to build the SOC, the incident response, all of the consulting and vir virtual CISO work. That's right. Niel: Which is when you and I met when I became a when I became a Brain Trace customer so many years ago. Yeah. Dean Sapp: That's right. That is correct. And so we did, you know, breach response and incident response. And with my my training with year sixteen years of being yelled at by lawyers, I learned a lot about what to say and how to say it and what to do and what not to do. So definitely, you know, learned a lot of of not to do things. But yeah, we built Brain Trace from from a very little IT services company to to a fairly large professional MSSP and manage SOC, manage STIM provider. And we built some really cool patented technology and SOPHOS loved it, so SOPHOS acquired the company. So three three plus years, probably the hardest years of my life because building startups are tough. Niel: Yeah. Yeah. Dean Sapp: Cause not only do have to do like, you know, all of your normal duties, you also have to you know, persuade people to trust you to buy your service. So Niel: Yeah, I remember I remember interacting with you back then and you you were always were you were always busy, right? Is that fair to say you were always busy, Dean? Dean Sapp: Always busy. Yeah, that was that was the I think my work day started at five o'clock in the morning in the office and wrapped up around six PM. But that was, you know, helping it with the kids at home and stuff. But I was that was a long, long day, week, month, year, years. So anyway, so after three years of Brain Trace, I decided to to become a CISO. at a a fintech company, did that for about a year. and I had another stop with another fintech for for a little less than a year. And then most recently I was reached out by another startup. I'm like, I thought I'm not gonna do this anymore. But I I agreed to take on another CISO role at a startup. Let's call it where I am today, Filevine. And so I've been at Filevine for eight years and Filevine's all things legal tech to run any kind of law firm, legal practice or or group. Niel: So Filevine's interesting to me. And just because I I do a lot of research and tracking of the use and the use cases for AI. And knowing you over these Filevine years, it's interesting to me how Filevine is using AI. So can we just talk for a few minutes? talk about, well, first of all, just generally. What's happening in terms of AI and security, both from the threat side and from the mitigation side. And then talk to me a little bit of how Filevine uses AI in general. I'm people may not be familiar with how Filevine is using AI. So let's start just talking about AI and security. Dean Sapp: Yes. Well I would say that AI has taken things, you know, say the creation of a zero day exploit and has exponentially increased the speed to market for an actual exploit. They do there's something called exploit chaining, where you may have a CVSS 3 around you know, a legacy crypto module. And then there'll be another module perhaps with a library that is I was end of life, then there's no hasn't been support on it, but it's you know, it's not public facing, so not really important. And and then so these AI agents will chain these different weaknesses together to then bypass the controls and ex and escalate the significance to maybe now it's a nine or a nine point five or or something like that. So so that's what we're seeing. with mythos and and now the the guardrailed version fable fable five fable five point one mythos five point one these tools are not just twenty percent or fifty percent or a hundred percent better they're two to three hundred percent better at finding unknown zero day exploits in the wild and of course that's why the government said you can use it just kidding pull it back Okay, try again. Nope, still finding fli flaws in the in the logic, pull it back. Okay, try again. You know, AI is very, very powerful. But we what we also learn from AI is it's often unpredictable. And if you give it a mission, it often will disregard future commands to to stop the mission. It's going to do the mission and it thinks that any any other communications between stop and start. or start and stop is is trying to to get it off course. So what we have learned is AI is incredibly obedient to the first command you give it. And after that, just get out of the way because it's gonna spin up a thousand sub agents and Niel: Yeah. Dean Sapp: like like we saw with the hugging face breach. I mean, you you have a competition, you you challenge the AI to win the competition The AI says, well, the answers are probably on hugging face. Let me send a thousand agents over there, break in, analyze all of the data, get the answers we need. We win. Yay! Happy dance. Right? I mean, that's that's AI. Think of AI as the most powerful seventeen year old you've ever met who is going to win the competition at all cost and disrespecting any law or or privacy. Niel: So, what does it look like from a mitigation side then? I had a conversation again. There are like a billion new AI cybersecurity companies, right? Maybe, maybe fewer than a billion, but not many more. How does somebody know how to move forward with cyber mitigation? Dean Sapp: Yeah, that's a great question. So we have built, I believe we're up to six or nine internal AI agents that we use. They're all very purpose built. The one most important thing that I've learned from AI is that you want a purpose built model, purpose built embeddings. You want AI orchestration, AI routing. And these allow you to properly secure it's Niel: Okay. Dean Sapp: that whole input sanitization, output sanitization. So you want to sanitize the prompting so that the prompting is not malicious. You Niel: Right. Dean Sapp: want to sanitize responses back and separate that back so the user can ask a prompt. We'll take a a legal use case. A user can ask a prompt, tell me. You know, what is my salary in the HRIS system? And the AI should be able to pull the data and say, Yeah, you're paid X amount. Where the prompting and filtering needs to be is you that same same intern should not be able to say what are the salaries of all of the other employees Niel: Right. Dean Sapp: in the firm. The AI probably knows, probably has access, but you have to have guardrails. That are tied to the user's authentication and authorization layers. So AAA has to be enforced for the AI agent. And so we've spent tremendous time putting guardrails in. you know, the hack the prompt database that's available on Hugging Face you know, 375,000 or it's probably far more than that now, known prompts that are designed to bypass filtering. Right. And if you're using a frontier model, you know, you're thinking, you know, Fable five point one or or some of these other models that are available, the latest versions of Kimi three dot three, I think, or you know, even the deep seeks and the quins of the world, those models are very, very efficient. They've, you know, frankly, they were stolen. I mean, if you if you look at the code, there there's open AI code under the hood. But you know, it's the Chinese have learned it's way cheaper to just steal really good technology and make Niel: yeah. Yeah. Dean Sapp: it better. So you ha you have to understand the nature of the model. You have understand what are the biases of that model, because every every model has bias and every and it's designed for a purpose. So you want to just make sure that if you're doing if we're using models, they are very purpose built and they are you're not allowing the model to train on client data. You have a zero retention, zero storage policy so that the actual Frontier model company does not keep the data. So after the prompting is completed, they should have a command that you can prove where it clears the memory. So there's no no residual of that particular prompt. But then you also need to have a prompt database, what's called a vector database. So you can store the computer readable data that's equivalent to a document. And you can then be able to have a real human verify the accuracy of that that data in the vector database, and then you can measure the accuracy of the responses from the different LLMs. So we've essentially built a router. That we can route to the most reliable, most accurate LLMs to get the data back. But you have to make sure you've got all the proper sanitizations and filtering and that whatever document gets embedded, you have it with a unique identifier so that you know the user who has access to that document. So as you're passing it through your orchestration engine, your AI routing to your embeddings model. And then out to the frontier models over a computer readable format that after it's processed, it's still maintaining that access layer. And then as it comes back and back into your vector database, back into your UI, that you've maintained that whole access control. It's very tricky to do if you don't have a very robust access control layer. But fortunately we spent three years prior to us building AI. We've been doing AI for three years now. So the three years private prior, we had built an extensible access control layer with permissions and overriding permissions and inheritances and things like that. Niel: Since you needed that anyway in the in the le in your for the for your law firm customers. Yeah, that makes sense. Dean Sapp: Absolutely, 'cause Chinese firewalls are real. Sometimes you'll have a firm that's so big it's representing both sides and they cannot see each other's data. And you can't let the AI see the different data either. Yep. Niel: interesting. Okay. that makes sense. Pierce it, yeah, that makes sense. So how do you think the world changed over the next few years from a cyber perspective? Dean Sapp: Well, if we take the Mozilla Firefox case study, you know, when they first used Mythos five, they were one of the like first forty companies. They they found literally hundreds and hundreds of zero days. So Niel: Yep. Yep. Dean Sapp: what I think this is going to do is it's going to speed up the the identification layer or tier of these vulnerabilities in all of the major cloud providers. So every browser provider. has already found and fixed, you know, generations of vulnerabilities that have been around for, you know, 20, 30, 40 years. Niel: Yeah. Dean Sapp: Right. The Linux kernel, you know, one of the things that Mythos did, some anthropic engineers, they they were so concerned with the number of vulnerabilities in the Linux kernel that they took a senior dev who understood Lena Linux extremely well, and they rewrote the Linux kernel in three days. using Mythos. Now Niel: Yeah. Dean Sapp: not every module, but the majority of the modules necessary to run a Linux system. And then they shared that data with the Linux Foundation. So that now the Linux Foundation can have a better operating system that's more secure. Microsoft's Niel: Yeah. Yeah. Dean Sapp: done the same thing. Google, Amazon, you know, so the benefit to to many consumers and many businesses Is that the platforms we're using every day are getting better at a faster rate than I think we've ever seen. I mean, they're really improving that. And now it's up to us as business owners to fix our code. I I have a friend, Trent Bond, really, really bright guy. he's a CISO, and he's like, PYC, patch your crap. Well, that that sort of is changing its on its head right now to just rewrite. Niel: Yeah, yeah. Dean Sapp: your crap because you can now rewrite the code so much faster and cleaner and say, look, Niel: Yep. Yeah. Dean Sapp: apply the OWASP top 10 for AI modules to this code, apply the OWASP top 10 for software best practices. By the way, make it highly available, highly redundant and scalable, and give all of those parameters to your your coding agent, to your cursor or your code rabbit or or Claude code, you name it, you give them to those providers, and now the output is actually remarkably good. Niel: Yeah, it's it's interesting you say that because for years we've never had the time or resources or money to really attack legacy application modernization. And now it becomes a reality. Like you said, do it the way you described it. I know what the outcomes need to be. Now I just to dramatically need to modernize, improve the quality of the code, eliminate the vulnerabilities, et cetera. So I think we're we're in up we're in a position now where we can eliminate this mountain of legacy. Dean Sapp: absolutely. Niel: stuff that has been lingering for years, that we've kind of crossed our fingers and hoped nothing bad would happen with it because we just didn't have the time to go fix it. Dean Sapp: Yeah, absolutely. Plus, it gives us an opportunity. I mean, C sharp, C++, they're not memory safe. So this is a chance to say, is there a new language like Rust, for example? Rust is memory safe, highly efficient, and a lot of the coding agents know how to write it. these AI coding agents know how to write in it. So be Niel: Yeah. Interesting. Dean Sapp: open to new opportunities. You know, I had this I have a senior app sec guy. Jesse Harris, super, super bright. we had a challenge with a legacy.NET app and some libraries and some code that we couldn't get priority to fix. I'm like, Jesse, you're a rockstar developer. H work with the agent and rewrite it. And he literally three or four days rewrote it Niel: Yeah. Yeah. Dean Sapp: and then gave it to QA. QA did all the regression testing, fixed all the vulnerabilities, and then the only legacy only waiting was waiting for our teams to roll it through all of the regression and make sure everything was good before they could roll into production. So that rewriting an app in three days is pretty remarkable. And that's the capability we're dealing with today. And just like my senior director of securities, Dr. Dale Rowe, he's got a he's got a doctorate in computer machine learning. So he's been using these same tools and we have transitioned from a buy first to a a build first environment Niel: Yeah. Build first. Yeah. Dean Sapp: because we can build custom security tooling for our exact environment and we can do it in weeks. So we yeah. Niel: Yeah, it's interesting. I you know, I I talked about for last couple of years I've been saying, are we getting closer to the point where I don't I might consider buying commercial software, but I'll try writing what I need first because the cost, the time and effort is so low compared to what it was three years ago. You'd never write your own CRM system. Well, maybe today I can. Anyway, it's an interesting time, Dean. All right. Last thing, let's wrap Dean Sapp: Yes. Yeah, it is. Niel: up. I've taken enough of your time. Talk to me about some of the outside of cyber, some of the things Filevine does with AI. From a from a law practice, from a legal perspective, what is the functionality that you use AI to deliver? Dean Sapp: Yeah, we we probably have twenty five or thirty use cases now, but I'll use a demand letter as an example. So law firms might have twenty or thirty years of historical demand letters that they have issued to insurance companies, opposing parties about p about specific types of matters. Well, now with AI, you can have allow the AI to access, you know, the twenty or thirty years of demand letters. And the whole the whole the most accurate way to d tell how if a demand letter is good is if it was accepted the first time. And Niel: Okay. Dean Sapp: by you know, because both parties are like, well, we'd be okay at this number. And, you know, because insurance companies know what they're gonna pay. They have historical data, not just from dealing with one law firm, but with dealing with thousands of law firms. So they already know where their comfort level is. And the goal is for the insurance for the the law firm defending someone or or getting a reward an award for someone who's been injured is to get as close to that number as possible the first time so that you save everyone time and money. And so we can help write demand letters in a day. And that's something that used to take, you know, six or eight weeks for legal researchers, for a managing shareholder, a couple of paralegals or staff and of council to kind of come up with their game plan and now you can have the AI generate that in twenty four hours based off of all of your historical data and then you could just make your fine-tuned tweaks. So it allows you to to serve a lot more people for with the same staffing, same resources. And so that's just one of many, many examples. Niel: So this is a philosophical question. And I'm thinking about cybersecurity as well. From the attacker perspective, and now from the attorney perspective, does it lower the cost, does it change the cost-benefit analysis? I used to always say, you know, the example I'm I'm hiking with my friend and we get attacked by a bear. I don't need to outrun the bear, I need to outrun my friend, right? And so I'd say, okay, I'm I that vulnerability, I don't think there's a huge impact from having it. I'm going to put a little bit of mitigation in front of it, but it's I'm not going to go crazy because from a cost-benefit perspective, the attack it's not worth the attacker's time and effort. Well, now the attacker's time and effort has shrunk dramatically. So does that cost-benefit equation change where I got to be worried about stuff I wasn't worried about before? And same, you know, I'm just thinking about that as an example with the with the demand letter. Do you do does an attorney now take on a client with a smaller demand because it's not going to take that much time and effort to get it resolved? It's just, it's interesting to me how this how the this level of lower cost changes the decisions on whether or not you do something. It's we live in interesting times, Dean. We really do. Dean Sapp: We do, and I think you're exactly right. It will mean that more people are given are able to get legal care because the barriers are lower now. It doesn't have to be a fifty thousand dollar settlement for them to even think about. Now they can they can put in the tooling and and use, you know, public data or data from Thompson Reuters or Westlaw or we even have a legal research product called Onbank. allow you put those those public sources in and it helps you understand what the norms are so that you can now help someone that you couldn't previously help. And so I think it's it is really improving the amount of legal care that's available to people who've been underserved for too long. Niel: Yeah, and this change cost benefit analysis, the same idea that I will now build my applications rather than buy them because the time and effort is so much lower. Anyway, Dean, it's always fascinating to talk to you. It really is. Dean Sapp: Thanks for having me, Niel Niel: I really appreciate it. I appreciate your time. we've spent some time today with Dean Sapp. Dean. ha is been a pioneer in cybersecurity. I'll just call it that, Dean. You've been a pioneer in multiple roles in multiple situations, including including building your own products and company to deliver cybersecurity. And now I'd say you're on the bleeding edge and leading edge with AI security and all its implications. So if you have a chance, get to know Dean. your life will be better for having known Dean. So with that we'll close and wrap up. Again, we appreciate it, Dean.