Season 4 Episode 2

Paul Mote on agentic AI and bespoke vulnerabilities

Guest:
Paul Mote
Published:
Duration:
27 min

Episode summary

Paul Mote, VP of solutions architects at Synack, doesn’t see agentic AI taking over for the most creative human hackers anytime soon. But he does believe in the potential for AI technology to radically disrupt penetration testing. 

In the latest episode of WE’RE IN!, Paul unpacks how agentic AI threats are forcing defenders to adapt by moving more quickly to discover and remediate critical vulnerabilities.

Are defensive AI agents really up to the task of finding the highest-impact software flaws in complex enterprise environments? Will agentic AI “helpers” leave security teams deluged by new CVEs? How do AI hallucinations change the game for the reliability of AI-only pentesting solutions?

Don’t miss Paul’s thought-provoking views on the future of AI for pentesting.

Listen to the episode

S4 E2 · Paul Mote on agentic AI and bespoke vulnerabilities WE'RE IN! Podcast · 27 min

Key takeaways

  1. AI agents miss bespoke vulnerabilities Paul expects AI to quickly match exposed, outdated SSH services to known CVEs, but not to spot an unreferenced web function and exploit it for remote code execution, a leap a human makes easily.
  2. AI slop deepens the vulnerability backlog Floods of polished but hallucinated AI vulnerability reports add noise for vulnerability management teams that are already underwater and struggle to find the signal.
  3. Exploitation often happens within a day Citing a VulnCheck report, Paul notes that 28 percent of 159 vulnerabilities exploited in the wild were weaponized within 24 hours of disclosure, faster than most change management allows.
  4. Automated pentesting is a tuned scan Most automated pentesting he has seen looks for known CVEs on infrastructure and leaves the rest of the attack surface out, making it closer to a vuln scan than a real pentest.
  5. Keep a human validation point Without human validation, AI output buries teams in content they do not trust. Paul points to the Target breach, where signal was missed in the noise.

Notable quotes

…the adversaries, they've always had the mantra of like, we just have to be right right once. Right? You have to be right a 100% of time.

…if you said, look at a website and look for a function that's not referenced by any API call in the web app that is publicly exposed and exploit it for remote code execution, it's not gonna be able to make that leap.

Because if we wait beyond twenty four hours, an adversary may already be in the network.

…the security industry had numerous cases over the last twenty years of burying people with information they don't trust.

Questions this episode answers

Can AI agents find complex vulnerabilities?

Not yet. Paul says adversaries are getting better at automating the boring stuff, but bespoke and complex vulnerabilities require logic and autonomy that current AI agents lack. Read this section

How fast are vulnerabilities exploited after disclosure?

Paul cites a VulnCheck report showing 159 CVEs exploited in the wild in one quarter, with 28 percent automated and weaponized within 24 hours of public disclosure. Read this section

Is automated pentesting the same as a pentest?

No. Paul says automated pentesting usually checks known CVEs tied to infrastructure and leaves most of the attack surface out, so it is a finer tuned scan rather than a real pentest. Read this section

Will AI help attackers or defenders more?

Paul expects attackers to benefit more over the next 24 months because defenders also have to overcome internal process, culture, and bureaucracy, while attackers only need to be right once. Read this section

What resources help someone get started in pentesting and AI?

For pentesting, he suggests Hack The Box, free content from Offensive Security, PortSwigger's academy, and vulnerable VMs. For AI, he recommends Anthropic's articles on agentic workflows. Read this section

About the guest

At the time of recording, Paul Mote was vice president of solutions architects at Synack, leading a team that works with clients to understand the business drivers behind their security testing programs. He describes the role as part consultant and part sales engineer. He credits a lifelong gaming background with shaping how he thinks about strategy and creative problem solving.

Full transcript

00:03

Intro

Carry nearly 80 gigs of data in

00:05

Intro

my head. You're in the mainframe.

00:07

Intro

It's eating the great entire system.

00:09

Intro

Access encoded. Gigabyte of RAM should do the trick. We're in.

00:15

Sponsor message

In. We're in. We're in. While Gen AI technology has enormous benefits, it also poses a threat if attacked by a malicious actor, which is why effective pen testing has never been more critical. Synack's penetration testing as a service platform connects your attack surface with the Synack Red Team, an elite community of skilled and vetted security researchers. Our experts are guided by OWASP's top 10 for AI LLMs to ensure your AI tech stays secure from the latest threats. To learn more, visit synack.com. That's synack.com.

00:51

Blake Thompson-Heuer

Hello, and welcome to WE'RE IN, a podcast that gets inside the brightest minds in cybersecurity. I'm your host, Blake Thompson-Heuer, joining me today is a fellow Synacker, Paul Mote, vice president of solutions architects here.

01:04

Paul Mote

Paul, welcome. Glad to have you here. It's an honor to be here, Blake. You know, longtime listener, first time caller. I really appreciate the, the slot this week.

01:12

Blake Thompson-Heuer

No. No. Paul, I have to admit, when I first joined Synack, coming up on four years ago now, actually, I didn't totally understand what a solutions architect did. I mean, it sounded good. Right? You're you're solving things. You're building things to solve things. I don't know. But what does it actually look like in in practice at an organization like Synack? Yeah. It's interesting.

01:31

Paul Mote

I think it's, a little bit of a Swiss army knife. Right? It's kind of consultant and sales engineering merged into one. I think our technical alignment is like the technical sales group, but I think we spend most of our time, you know, talking to clients, understanding what are they trying to solve with their security testing program? Why are they trying to do it this way? What what are the real business drivers behind the testing that they're doing, not just the fact of doing security, and then making sure that whatever we we ultimately put in front of them maps into what they need to do to to guarantee the outcomes they're looking for and make sure that they're gonna be set up for success.

01:59

Blake Thompson-Heuer

Given that you're on the technical side of the house here a little bit with with some of, how you navigate the challenges that customers are facing, I... I'd be curious to hear your thoughts on AI. And and normally, I do wait and cover a couple questions before jumping right into the AI buzzword, but, I guess listeners can check that off their bingo card right away because it it really is important, and it... It's especially fascinating how generative AI technology and just the rapid advancements we've seen in the last couple years. How has that changed the way that you approach your day to day work?

02:28

Paul Mote

Yeah. It's funny. I think you nailed it. Right? I think you can't have any conversation without at least talking about AI for a good chunk of it. I think AI is the the technology that's gonna change the the the space as we know it for everything that we do. So I think for me, you know, I I look at it and realized very early on, it's a great companion for brainstorming. It's great for research. It helps you kinda fill in the gaps in areas that you may have in terms of your own skill set as long as you're honest with what your gaps and your skill set are. I actually had a friend that said, I wanna learn AI. What what should I do? I said, we should ask AI how to learn AI and have it build you a a plan. And they're like, I didn't even think of doing that. And I think that's the fundamental shift of it's just becoming like the new Google. Right? You can't... Rather than Googling it, usually you're gonna leverage AI to go help you do the thing and figure out the thing. And naturally, with the introduction of, like, agentic workflows and everything else, everything kinda changes in terms of how to leverage that capability.

03:21

Paul Mote

So I think it just... It's part of your daily life for for everything.

03:24

Blake Thompson-Heuer

Well, we'll get to the agentic piece soon, but I I did wanna flip to the adversarial side of the house. How do you see US foes using or potentially using AI to retool their offensive operations? Right? I I feel like we hear a lot about the AI enabled attack, and I've even had some conversations with other podcast guests this season on that. And I... Sometimes in practice, feel like it's... You you... We're not always seeing generative AI leveraged. And sure, it might be improving phishing email quality and whatnot. But where does the threat sit as we speak, and where might it be evolving?

03:58

Paul Mote

Yeah. I think, you know, with everything good, there's always the opportunity to use that same good for bad. So when you're sitting in your chair and you realize you get answers faster, you can build these things faster. You can have the AI look at your code and help you fill in the areas that you're weak on your code or write better code, integrate solutions that you didn't think could be integrated because AI can help you do that. All of those same capabilities earn the adversary seat. Right? They have the same subscriptions that we all have. They have the same capabilities we saw with the Chinese version of ChatGPT that came out. Right? Be able to very rapidly spin up and synthesize data and be able to to execute more quickly in terms of learning learning from larger datasets. Adversaries are gonna ultimately be able to leverage the same same capability. So researching organizations to understand what they're doing, looking at their attack surface, analyzing what is in their attack surface, being able to figure out how to pair the right attacks against that attack surface.

04:49

Paul Mote

It's gonna help exponentially increase their capabilities, and we have to be ready for the fact that they're gonna be able to to be better at finding our holes faster. And ultimately, that that rapid escalation of of them being able to look at information and get to us faster and be successful is something we're gonna have to come to terms with and really have a major shift in terms of how organizations handle what their vulnerability and attack surface exposures look like.

05:12

Blake Thompson-Heuer

No. That's that's definitely a a sobering thought, the potential for these adversaries to really scale up their operations and leverage AI. I mean, we see... You mentioned in your day to day work, I certainly use AI as well. Right? It is... It does help with that scale piece of being able to produce more, accomplish more. Are agents really up to the task, though, of finding those gaps that you described in those holes in people's networks? Are they really finding and exploiting critical vulnerabilities? As things stand, perhaps they're a little bit better at carrying out these more discrete concrete one specific task type operations. And, you know, some of the vulnerabilities certainly that you deal with, Paul, are are... Can be really complex pathways that take multiple hops and steps and almost logical leaps to to to arrive at. Are we gonna see agentic AI better or grow better at picking off some of those more complex ones? Are they gonna stick to the low hanging fruit?

06:03

Paul Mote

I think

06:04

Paul Mote

I think eventually, we we will see AI start to be more successful in the things that we think are too complicated for the AI agents to handle today. I do think the reality of our current state is, adversaries are getting better at automating the boring stuff. So just like we should be automating the boring stuff, they're getting better. The larger the data data pool that they can learn from, the more successful they're gonna be in it. Right? So the more bespoke and complex it is, it requires a lot more logic to be inserted and a lot more autonomy of the AI to make its own decisions, and it just doesn't have that. Right? So if I'm looking for SSH ports that are exposed, that are using an old version so that I can then pair a known CVE to it, I would expect AI to be pretty quick and be able to do that. If an AI agent is focused on SSH and that SSH exploitation, it'll probably do pretty good. But if you said, look at a website and look for a note and look for a function that's not referenced by any API call in the web app that is publicly exposed and exploit it for remote code execution, it's not gonna be able to make that leap.

07:01

Paul Mote

A human will make it pretty easily when they're going through it. Oh, that's weird. And thought process research. Oh, that's weird. Thought process research. Oh, I found this thing. Now I'm in. I think there's a big gap between AI being able to do that compared to where it is currently today.

07:14

Blake Thompson-Heuer

I'm glad you mentioned CVEs because I feel like that also plays into some of the discussion around around scanning technology and automation in general. There's a big difference, obviously, between known CVEs and unknown CVEs. Where does that factor into the AI discussion? Right? Figuring out... A... Or AI, are you gonna see them dropping new CVEs easily? I think there might

07:36

Paul Mote

be some low hanging fruit stuff that ultimately feeds into that. I think the challenge of... Like, when you think of CVE, right, CVE stands for common vulnerability exposure. So that's the... Just in case folks aren't tracking CVEs, that's what it stands for. And really that process is I found a problem in a piece of software, and I've reported it to a naming authority. And they assign a value to it so that the intelligence associated with it can be shared with everyone from a defensive and scanning perspective. So how do you patch it? How do you scan for it? How do you look for all those things? I think if we get to a point where AI is finding and submitting what it thinks a CVE will be, it'll just make the pile of vulnerabilities, which is already going up into the right, astronomically uncontrollable. I don't think that's really a good a good use case. I think it would get out of control very quickly. I would think that AI outputs for low hanging fruit techniques like looking for exposed file types on web apps, like information disclosures.

08:28

Paul Mote

I think that will... That should help with creating some additional value in terms of finding vulnerabilities at scale.

08:37

Blake Thompson-Heuer

Glad you mentioned that noise point, right, of our AI systems. And I guess now we're flipping a little bit to the defensive side here. Are they gonna be so good at finding potential things that they just deluge defensive teams with a bunch of junk? Right? And these AI reports can be often really well written too. I'm... There are gonna be very polished looking. But then if you look under the hood and realize, okay, this vulnerability isn't actually exploitable or the risk is, you know, pretty acceptable for what's happening here. And suddenly, you have hundreds and hundreds of these reports to sift through. Is that really helping anybody, or are people just gonna be drowning in AI vulnerability slop, I'll call it?

09:15

Paul Mote

Yeah. I think that the... There was a open source program that talked about a program they were running, and they were getting... Since AI was released, they're getting a a massive flood of vulnerability reports to their bug bounty program. And they said it's all slop. It's a 100% low quality, unactionable. The AI outputs are hallucinating. They're pulling in snippets from Reddit that aren't actually inside of the the... Like, in the POCs that they're giving us or not even available on our website. And it's... I think that is the... That is a major risk. And it... And it's a risk that amplifies the existing problem, which is our vulnerability management teams are already underwater. They simply can't handle the stack of known vulnerabilities they have today. They don't know how to prioritize them, but they also don't know how to get through that stack because it just keeps getting higher and higher and higher. Management processes are difficult. So I I think it could definitely perpetuate a lot of the noise and make it even harder for those teams to figure out where the signal is. Like, what do I actually need to take action on to keep us out of the news, protect our customers, to protect our shareholders?

10:10

Paul Mote

I think that's a it's a major threat to that to that in space.

10:13

Blake Thompson-Heuer

We hear a lot about time it takes adversaries to exploit vulnerabilities in the wild. Right? I love that phrase in the wild. It sounds like David Attenborough is narrating advanced persistent threat activities. But... So out in the wild, we're seeing this time to exploit go down. Maybe it's kind of leisurely gone from weeks to days. What will that time to exploit look like a year from now in the AI era?

10:37

Paul Mote

Yeah. I think it's it's probably one of the biggest threats that organizations face today is is how fast that that window is shrinking. Right? VulnCheck actually had an amazing report that they released in q one of of this year. They highlighted that a 159 vulnerabilities, meaning that there had CVEs associated with them, were exploited in the wild. And for those that aren't tracking what in the wild means, it means that an adversary, somebody who's not authorized to attack an organization is using it for some level of gain. It could be a nation state, be activist, could be something else. But they create a breach. Right? They break an organization using those vulnerabilities. The interesting thing of those those vulnerabilities that 28% of them were automated and weaponized within twenty four hours of public disclosure. So you kinda sit there and imagine, you know, Blake, you and I are running an organization. We have a thousand systems that we're managing, hundreds of people that are working on this stuff. And all of a sudden, a vulnerability drops in the news.

11:29

Paul Mote

And somehow we have to see that that vulnerability drops, be able to ingest that information, and figure out do we have an exposure, and then figure out how to solve it in less than twenty four hours. Because if we wait beyond twenty four hours, an adversary may already be in the network. Wow. And you take that and say, well, how do I get through change management? How do I make sure that a patch, if it's if it's even available, is going to not break my system and cause downtime as I'm trying to defend my network. All of those things are gonna change our our landscape dramatically.

12:00

Blake Thompson-Heuer

Well, and you've been really good, Paul, about defining some of the acronyms that we like to toss around in the cybersecurity industry, right, CVE. And we have another one that we like to use at Synack, the mean time to remediate or MTTR. I don't know exactly offhand what the what the number is with the averages, but, you know, you're not seeing many folks patch with the kind of speed you just described, right, that that that twenty four hour window that perhaps is only shortening from here on out, are we falling behind?

12:28

Paul Mote

I think as an industry, that is a... It is a core problem. Even in our our outputs. Right? So we have the benefit. We have the Synack Red Team. When we give a report to a client, we give it to them as we find it, and it includes the steps reproduce, recommend fix action. All of it is included in there with a full exploit chain of how we did it. So they get it real time during the test. Even then, organizations still take a while to get through their change management processes to review that information. So we can say it's real. It's actionable. I need to fix this because a human being won and it's confirmed. It still takes days, weeks to get through those internal processes. And so knowing that that is already happening today, you're already getting that level of high fidelity intelligence. It's definitely gonna require a lot of reworking so that organizations can get intel and act on an intel a lot faster. Right? They're gonna need... How do I get the the signal faster? And then how do I act on that signal faster? Because I won't have the the benefit of it taking a long time.

13:19

Paul Mote

Right? Otherwise, I'm just gonna have to deal with the fallout of whatever comes if I'm if I'm late to the the party.

13:25

Blake Thompson-Heuer

Well, and presumably, that's gonna necessitate leveraging AI as a defender to some extent to speed some of those processes up, I would imagine. But there's an interesting cultural element to this too. Right? We... You know, you talked about using AI in your own workflows. I think there's certainly a lot of pressure and good reason to turn to some of these AI tools. But certainly in my space, you know, in the communications front and, former journalist, I've seen there's a lot of skepticism, lot of reticence for the uptake of AI tooling. Right? People don't necessarily trust that the AI is gonna deliver accuracy or performance that they need. So you still have potentially, you know, a good chunk of people not turning to these tools that could accelerate some of these, you know, time it takes to remediate vulnerabilities and whatnot. I guess, what are you observing working with a lot of a lot of clients and Synack, you know, prospects on that cultural side.

14:16

Blake Thompson-Heuer

Are people ready to embrace AI and apply it across their securities function? Or is there still a lot of reticence? What have you been noticing?

14:24

Paul Mote

Yeah. I think it's interesting. There's there's a weird divergence in terms of how people feel about AI. All of the security teams are now having to respond to the fact that AI is already in their products that they provide to their customers, and they may not even realize that it was there, but it's there. Right? If any application that had a chatbot, it's probably run on AI now on the back end. And that one of the major challenges on the defensive side is that the vulnerability classes of AI are different. They're not the traditional vulnerabilities that you would experience. It could be things like, you know, could I get, you know, bias testing? Like, will will the application say something offensive to my customers? I need to make sure that the user who's who's attacking that chatbot, since I don't really get insight into what it's doing and what decisions it's making and how it's making those decisions, I need to make sure it doesn't tell the the end user how to do something illegal. Right? I don't I don't need to be the source of how they got information, to do something illegal.

15:14

Paul Mote

Right? Which we... I think we had a blog post we wrote about somebody who was able to jailbreak one of the AI chatbots that we were testing and ultimately get it to feed them some very interesting information. So there's that. And on the Defender side, there's still an inherent need to have a human in the loop. Right? I think that's the number one thing that folks were kinda looking for. It goes back to like the security orchestration days where they were trying to say, well, if an email comes in as it has an attachment, automatic do... They do all these things all the way to quarantine. And everyone said, no, no, no, no. Like, we can't we can't trust computers to go do that for us. So I think the... In terms of where to integrate AI, how much to trust AI, that's still a a area that the security teams are having to to look into and ultimately figure out what they're comfortable doing. But you see it everywhere, whether it be, you know, Microsoft Sentinel, right, integrating AI for Copilot, all the way to Splunk integrating Copilot. And even Atlassian now includes Copilot...

16:05

Paul Mote

Some kind of AI for their developer workflows. I mean, AI is is everywhere, part of every tool that you're working. So how do you integrate? How do you how do you adapt it?

16:14

Blake Thompson-Heuer

Well, just doubling down on that, I guess it's no surprise you hear one of the top AI companies, Anthropic, their security leader CISO, Jason Clinton, told Axios in the spring that he expects AI powered virtual employees to start roaming around corporate networks starting really by next year at the latest. That's, you know, 2026 right around the corner. Are we gonna start to see AI powered virtual pentesters?

16:45

Paul Mote

Yeah. I I think every every industry, if it's technical, has has an element of how an AI agent would fit in and do some level of function. Right? We've kinda seen it on the developer side already where they're using AI to be most of the junior developers and having senior developers review the the code of junior developers, which are really just AI agents. That's all gonna play into, you know, how do we secure the network whenever the majority of traffic across the network and across all of our toolbases are not humans. And we don't necessarily have insight into all of the pieces and decisions that they're making and how they're making those decisions. So there's a there's a lot of interesting dynamics and and challenges associated with that. I wouldn't be surprised if we're able to do that in terms of going back to the Anthropic comment. I do think organizations will probably be a little bit slower on critical functions. They probably won't just implement that right away.

17:37

Blake Thompson-Heuer

Right. That speaks to the importance of the point that you raise, a human in the loop, to kind of trust but verify a little bit the AI outputs. That makes a lot of sense to me because I'm even just thinking, you know, Synack fought this battle years ago, right, of demonstrating to organizations that they can and should trust ethical vetted hackers from around the world to, you know, test their networks and give a crack at it. That was quite a radical concept at some point. Right? And now it's it's pretty well accepted that that's the best security practice, getting some of that adversarial perspective. It's not a... It's not an uphill battle culturally to have those conversations with CISOs and, you know, security leaders. But I imagine if you're talking about just throwing a bunch of AI agents at critical assets when, you know, even the AI companies themselves admit that there are some black box issues and not necessarily gonna be be able to dissect with granularity what these AI products are doing with their decision making, like how they're arriving at certain decisions.

18:33

Blake Thompson-Heuer

I don't know. That would make me squirm a little bit as a security as a security leader. How do

18:37

Paul Mote

you win their trust? Yeah. Absolutely. And that plays in the offensive security space too. Right? There... I've seen a lot of of fluff out there in terms of, like, you know, this agent can go do all these things, and you kinda dig into it a little bit, and you're like, I don't exactly sure. I I I would a 100% buy into what they're doing. I think there's there's still a element on the offensive security space of the the low hanging fruit stuff should be... Can be automated and even even we're seeing a ton of success there, right, in terms of like what an AI can do and look for. But I still think for a lot of these, you need a human validation point. Otherwise, the noise factor becomes exponential, and you just bury teams and stuff and content that they don't trust. And the security industry had numerous cases over the last twenty years of burying people with information they don't trust. And all that does is create a culture where people don't take action on the things they need to, or they miss the the signals that they need to be looking at and have horrible outcomes right now.

19:30

Paul Mote

I think that the Target breach was a good example of that, where they had signal, but their team wasn't looking at the signal because they were buried in the noise. And so had they seen the signal earlier, probably would have been a... Still would have been a breach, but it probably wouldn't have been as bad. I think that's one of the challenges of those... The AI outputs.

19:44

Blake Thompson-Heuer

You mentioned, you know, obviously, there's an element of autonomous AI function here. Right? The the AI kind of doing things on their own, making certain decisions, following certain attack paths. Is that the same as, like... To to play Gartner here for a second and make some definitions. Is that the same as automated pentesting? Is there... Is that the same thing as scanning? Like, where... How do we define this? How do we categorize this? And are there certain things that are better or worse when you're really looking at, you know, which tool... Which pen testing tools might be worth their salt?

20:16

Paul Mote

Yeah. I think it's it's interesting. I think there's a lot of there's a lot of misinformation, confusion around the notion of automated pen testing. I think if you if you talk to any pentester worth their salt and say, hey, automate this pentest flow, they'll go, I can automate the scanning, and the scanning can look for the known knowns, but I don't trust any of the outputs. It's all suspected until I can take action on it. I gotta prove that it's actionable. Otherwise, if I send this all to the end customer, they'll go, yeah. No. No. We've already got all these scan results from Tenable, Qualys, Rapid7. You Yeah. Yeah. Yeah. You just kinda go down that So I think from the automated pen testing side, usually what what I've seen in in my exposure to it from organizations who've leveraged it, it's usually looking at known CVEs tied to, like, infrastructure and kinda leaving the rest of the entire attack surface out of the picture. So it's just kind of like a finer tuned vuln scan compared to, like, a real pen testing capability.

21:07

Blake Thompson-Heuer

So fundamentally, that can only... This one piece piece of the puzzle is what you're describing. It's not giving that kind of holistic, top to bottom, you know, pentesting all your potentially exposed assets. That makes that makes sense. And I think you're going to see I mean, certainly we're seeing it with AI, right? Like the claims flying around everywhere. It's difficult to sort out what's real and what's not when when it seems like we're in an era where AI can do anything and everything. How do you get to the bottom of what's actually, what it's actually capable of?

21:35

Paul Mote

It creates a lot of confusion. Right? It'd be like calling an Nmap scan with a plug in, you know, a vuln scan. It's like, nope. It's not. It's a port scan. It's a smarter support scan, and then it's like, can you go up? It's like a vuln scan. It's a vuln scan. An automated pen test is like a vuln scan with, like, a little more little more intelligence into it, and then you get pen testing and red teaming and everything. It goes just well beyond those capabilities.

21:54

Blake Thompson-Heuer

One question I like to ask just, again, centered on so much that we've talked about with AI, and I love hearing your thoughts on this. It's it's it's a really important area and one I'm sure a lot of our listeners are keenly engaged in, is just at a high level, do you see AI helping defenders or attackers more, say, in the next year? I know I know it's hard to predict with these sorts of things, and and, you know, predictions are tough as they say famously. But where do you see AI helping more?

22:21

Paul Mote

I think it's on the attack side. Because I think that on the defense side, it's not just about the technical capability. It's also about the internal process and the culture. There's a lot to overcome beyond technology for the defenders. Whereas the adversaries, they've always had the mantra of like, we just have to be right right once. Right? You have to be right a 100% of time. We just have to find one opening and then we're in. And so... Yeah. Yeah. Pun intended.

22:43

Blake Thompson-Heuer

Cue the cue the music we're in. We're in.

22:47

Paul Mote

I love it. Right? And so I think their their capabilities are growing exponentially, and they're gonna be able to attack targets faster. And they don't have the internal bureaucracy or politics or, you know, any of those internal key processes they're gonna have to go through to take action on it. And so I think it'll benefit the attacker drastically more over the next twenty four months compared to the defenders.

23:08

Blake Thompson-Heuer

Finally, for somebody who's really interested in this and listening to this podcast, say, and is just feeling a little overwhelmed around a lot of fancy terms and, you know, pentesting, processes, if if somebody were just looking to get started, right, in AI or in in pentesting even, what resources would you recommend? Where where should they go?

23:27

Paul Mote

Great. So if you're on the on the pentesting side, there's a lot of materials out there now. It's kinda... It's very popular to get into the offensive security space. You're always learning. There's always niches to go after and be the expert. Like, that one niche, you could do, like, hack the box. There's a lot of free online resources from offensive security and the the folks over at Crepe Burt, PortSwigger. Right? They have a bunch of academies, things you can go through to learn how to hack various things. There's various... Like VulnHub has a bunch of vulnerable VMs that you can attack and learn in your own home environments. There's a lot of cool capabilities to kinda get in and figure out, like, is this for me? And like how deep down the rabbit hole I wanna go. But if you're gonna go in the the pentesting space, you know that you've gotta go deep into how everything works. Right? You gotta understand how systems work, how protocols work, all of that fun stuff. I would dig in there. On the AI side, I'm actually a huge fan of Anthropic's content. So, you know, personal subscriber to Anthropic.

24:18

Paul Mote

I love their their agentic workflow articles out there that really define, like, how an agentic workflow can be and all the different variations and flavors and how each agent can be defined. I think that's a great place to start to kinda open up the eye... Your ideas of, like, how could AI automate an end to end workflow of anything you're doing, security or otherwise. How would you build agents? How would you go through it? How do you think about the architectures that can be implemented? And then from an offensive standpoint, I look at that same thing and go, well, where can I break that chain depending on how you've decided to implement your your AI workflow? Because there's potential for escape down that process. So those are the two place... Those are the areas I'd probably start.

24:56

Blake Thompson-Heuer

No. That makes a lot of sense. And I... I've seen, you know, Anthropic's big on the transparency piece. So they really lay it all out there. I've seen some of their blogs and content that's, like, extremely... Almost damaging to the AI industry when you look at it, like AI trying to blackmail people and things. And they lay it all out there because I think they realize that it's important to reckon with the ups and downs of AI technology, and, it's something that certainly the conversation will continue. And, and, Paul, I I really appreciate you joining for this conversation. I think it's an important one and and certainly an interesting one. Now finally, before we, we part ways here and get back to our day jobs, there is a question we like to ask of all our guests, which is what's something we wouldn't know about you just by looking at your LinkedIn profile?

25:38

Paul Mote

Yeah. That's a good one. What can I say that wouldn't be used against me in social engineering campaign in

25:43

Blake Thompson-Heuer

the future? Thinking about that.

25:45

Paul Mote

Of course. Of course. You gotta you gotta think about it that way. I think most most people who who know me know that I actually have a a huge gaming background, and so I grew up being an avid gamer fueled by ADHD. Just, you know, naturally latch onto it for hours and hours and hours. But that's probably one thing you you wouldn't find is that I'm a I'm a huge gamer. But it also plays out, as you can probably tell, in a lot of the ways I I think of strategies and getting creative when you run into problems and whatnot. So that's a it's a a skill and passion and addiction that has been very helpful to me throughout my career.

26:16

Blake Thompson-Heuer

Well, tell tell me more. If I'm gonna log on to Steam later, where am I gonna see where am I gonna see a plane? And the lot... Which lobbies are you popping into? Oh, I I gotta be secret. I gotta be secret.

26:24

Paul Mote

Alright. I

26:26

Blake Thompson-Heuer

I won't... I'll I'll look out for Paul underscore underscore Mote seventy seven somewhere, and and I'll know it's you.

Enjoyed this episode? Browse all WE'RE IN! episodes →