Mackenzie Jackson: Hi everyone. Welcome to another episode of The Secure Disclosure. I'm here with Norman. Norman, thanks so much for coming and chatting with me today.
Norman Menz: Thanks, Mackenzie. It's a pleasure. I really enjoyed the pod and frankly, I'm looking forward to the "What Would You Rather?"
Mackenzie Jackson: OK, I'll have to make spicy ones for you there.
Norman Menz: Please.
Mackenzie Jackson: So one of the questions that I just wanted to kind of roll with is just, what exactly does Flare do? What is the problem that existed, that exists now, that was created to solve?
Norman Menz: So in a nutshell, Flare is the only identity for CTI provider. Sounds pretty straightforward. What does that mean? I'll kind of unpack it a little bit. We started the business back in 2017, really as a traditional cyber threat intelligence provider. So focused initially primarily on the dark web. So any dark web marketplace or forum that's involved in financial fraud, cyber fraud, et cetera, we would have access to and take all of that data, index it, make it searchable. And ultimately, at the end of the day, the reason we founded the platform is we wanted to remove the need for people to have years of cybersecurity experience to navigate the complexity of not only dark web services and Tor, but also clear web and other kind of grey web services and bring that power together in anyone's fingertips, whether they're a SOC analyst, whether they're a pen tester, red teamer, or threat intelligence, threat hunter, et cetera, and really make it easy for people. So within seconds, you can conduct very complex searches across not only the dark web, which is where we started, but the clear web and then other things like Telegram, GitHub, et cetera. But anywhere where you have sensitive information exposed about an organization, whether it's intellectual property, API keys, identities being the big focus for us, we wanted to make it really easy for people to get at that.
Mackenzie Jackson: There's a lot of questions there. I've been asked a lot and I've never been able to answer this well, so I want to ask you is what exactly is the dark web for people, like why and how does that exist and why does it exist?
Norman Menz: So the dark web, I'm sure many of your listeners kind of knew the history of it, it's really created off of the The Onion Router network or Tor. But essentially what the dark web is, it's kind of an abuse of the Tor network for nefarious purposes. That's generally how we describe it. It's the unindexed part of the internet, so to speak. And there's the classic separation of the dark web, which is things that run over the Tor network. Then you have what would be considered the deep web, but those are things on the internet, but typically behind some type of a login wall. This could be something like Facebook or something like that. And then obviously the clear web, which are the things that are open. But the dark web, typically, at least in a cyber perspective, these are marketplaces and forums, typically run by the threat actors or threat actor groups to exchange illicit information, trade credentials, things of that nature, provide initial access to organizations, or even team up on targeting organizations.
Mackenzie Jackson: What's like the legitimate use case on it? As in, like The Onion, it was created for a reason that wasn't malicious.
Norman Menz: So it was created by the US government as a secure way to exchange information across the internet. And it's even used today for very legitimate reasons. You have people in countries, for example, where their laws don't necessarily permit free exchange of information, et cetera. So you can have reporters and other journalists, et cetera, who will use it to securely exchange information. The thing that's interesting about the dark web is, and we talked about Telegram a little bit, is Telegram is starting to take the place of that. So you have a sufficient level of anonymity.
Mackenzie Menz: It's not. It's not the same because it is somewhat attributable and the one thing about Tor that's always difficult is very slow and Telegram removes that. Your large file capabilities you have bots and other types of automation that don't necessarily exist.
Mackenzie Jackson: Yeah. I was, I was going to want to kind of go into that too because it is interesting, is that like, yeah, the Telegram has kind of replaced a lot of this, you know, there. Yeah, there's because they almost act as forums as chat rooms in there. Does Telegram work in the same way in terms of when I think of the dark web, I'm thinking of just in my role, in my experience, hacker forums that are typically exchanging stolen information, selling stolen information, that type of thing. Does Telegram also fit into that? I mean, I know there's a lot of a kind of illicit activity on there, but there's that also, OK, that marketplace.
Norman Menz: Absolutely. I mean, almost every activity that you will see happening on the dark web, you'll see happening on Telegram, and actually then some. So you know, the thing that's really interesting as we talked about the automation that you have in Telegram, you have bots. We see Telegram very often actively being used for command and control infrastructure. We're finding active phishing kits, and if you kind of unpack those phishing kits that are out there, you'll see the command and control are Telegram channels. And so it's, you know, I talked before about the speed, like I can't overemphasize how important it is to be able to do large file transfers and things of that nature that is just almost impossible on the dark web. Yeah. And it's kind of really the new movement. I think it's around 2021 where we really started to see a significant movement of that activity over to Telegram.
Mackenzie Jackson: Something that I've always couldn't wrap my head around it specifically about you and it's kind of the other players in the space, the dark web is very slow and you all are able to do some pretty large processing of information on that. How does that work? Because those two things feel completely in conflict with each other, that you're able to access all this information at very fast speeds on a very, very slow network.
Norman Menz: So I won't go into all of it because some of it is not only proprietary trade secret, but also, obviously kind of given the adversary. So generally speaking, what we look to do is we look to archive that information and bring it back into our data store so that we can do feature extractions, we can do indexing of it. We can start to bring like graph databases and linkages across the different elements so that our customers can pivot across that data, search it very easily, do both simple searches as well as very complex searches. But more importantly, give our customers the ability to continuously monitor using those search or query terms so that anytime there's an event that matches that, they can get immediate alerting on that. And then where possible, when you can do things like remediation and prevention, obviously that's the most beneficial capability there.
Mackenzie Jackson: Yeah, yeah, super interesting. There's so much that I want to get into this podcast, so I have to try and get off this topic. But I was curious at the start that you mentioned that you monitor a lot of places and the dark web, that you know, your network. You also mentioned Telegram and then you also put GitHub in there. I was curious about the GitHub moment. What are you looking for on GitHub? Because it doesn't seem to me to be a place that this nefarious activity is happening, but maybe it is. But there is just a lot happening on GitHub. So perhaps that's behind it.
Norman Menz: I mean, sometimes I think about that when you think about the movie Jurassic Park with Jeff Goldblum, he says like nature finds a way. I always say crime finds a way. So when you look at GitHub, it's both nefarious and accidental, and so to touch on nefarious very quickly, it's often times used for command and control. Right, the Shalott examples. And so you there'll be a couple of things. You'll see it being used as command and control. So these are giving instructions to different agents that exist about where to go next, where to send files, et cetera. There then you also see, we'll see a lot of the supply chain attack where we're injecting malicious code into packages and things of that nature that maybe in public repositories. And then the thing that I think a lot of our customers are really concerned about is accidental disclosure. So these are when people make mistakes and in these particular cases you've got developers or third party that accidentally commit code, API keys, cookies, et cetera, sensitive system configuration with a commit. So what we essentially do is we're monitoring the public repositories. Anytime we see commits that are related to our customers and the things that are important to them, we'll go ahead and clone that repository and then we'll scan it. We'll look for sensitive information and then alert them to that and then they can take the appropriate action around that.
Mackenzie Jackson: Yeah. And you did also touch on like what you're looking for. And the obvious one for me is like credentials, right? That we all know that credentials get stolen on the dark web, but there's also lots of other stuff like GitHub's whole source code was recently for sale on the dark web and other areas like that. What are you, what are you actually looking for? And one of the reasons why I'm so curious about this is that finding API keys, like we could go into a whole rabbit hole of why this is hard, but at least there are things that you can look for if they're like prefixed or something like that. PII very difficult to find within grips or within search parameters. So like, what are what are you actually kind of looking for in these spaces? And I'm not going to ask you how you find them because I know you can't tell me, but I'm very curious as to know what you are looking for.
Norman Menz: So we run what we generically call like different detection engines that look for this information. And if we're still on the topic of GitHub, it's not impossible to imagine a scenario where you've got, perhaps you have a software engineer that's working on a project related to clinical data and sample data gets included in that. So that would be something that would be an example of where you would see some PHI in this example that would be included. There's a lot of other things that are difficult, but then there's some things that are very obvious where you can do pattern matching on them. And often times when you match against one pattern, for example, could be Social Security number or things of that nature, you'll find other related things very quickly. And then obviously today we've got the benefit of LLMs. So it's very easy to run a lot of this data when it is interesting and relevant through different models to identify PII.
Mackenzie Jackson: Yeah, yeah. That I think that is like a great use because I do want to get into LLMs as well. One thing that is curious to me though, is organizations that are looking for this type of information, how does that element fit in, in terms of like, how do you even know what to look for? So I'm just going to assume, and they said there's an organization that wants to know if their sensitive information, PII is leaked on the dark web. But like connecting those dots seems very ambiguous.
Norman Menz: So sometimes it's fundamentally easy. In many instances, you can just start with an apex domain for a customer, right, for the organization and you're going to search on that and then you're going to find emails associated with that domain. You're going to find hosts associated with that domain. That is usually the initial entry point and then you're pivoting. A lot of times customers will know what's important to them or what's sensitive. For example, if you go back to the GitHub example, just for a moment, like some customers will intentionally use honey tokens or honey functions, for example, if this type of pattern is ever seen outside of their four walls, they want to alert on that. So there's different things that more mature customers may do to inject things into the information in case it is leaked. But more often than not, you're starting with, you know, maybe there's a special project a customer is working on. They're still sensitive. So they'll be searching for those as kind of keywords and other types of information. We have the ability to do complex queries with regex and things of that nature and where you can join things together and really get very specific about a customer that's talking about accidental disclosure. If we talk about the other side where we've got organizations being targeted, you can simply put together things, various keywords about the entity itself, whether it's a domain name, maybe it's its name, maybe it's its product name. And then other things like fraud or if you're if online retailer gaming, you could talk about cheat codes and things of that nature. So you start to see information bubble up very quickly.
Mackenzie Jackson: I'm very curious. We don't want to ask you this too and kind of get into that. Like it's always a joke. It's like, how long can we last for a podcast before we start talking about AI? How, how is it? Like I think there's a lot of ways to kind of approach this question too, but like I'm very curious. It's like how, how has AI changed this system? You mentioned briefly about some advantages that you have in detection of certain things with AI and I can see that how is it kind of changed on the adversarial sides and kind of given them new abilities that we maybe haven't seen before?
Norman Menz: I mean, I think this is the topic du jour right now for everybody. First of all, I think it really boils down to like, who's going to benefit the most from this? Is it going to be the attackers or defenders in terms of leveraging AI and generative AI, both to defend and to commit attacks? I think it was just yesterday news started coming out that there were attackers using LLMs to do phishing attacks against like very high profile hedge funds, right? And this happened in late July. And then the news is just launching now. I mean, we see examples of it all the time. We see examples of malware that's being generated by AI that's being used, especially in the info stealer or from a phishing perspective. So the attackers are adopting as quickly as you know, we've got our friends sitting next to us at the office, five coding from business application. We've got attackers doing the same and we're seeing it actively on a weekly basis.
Mackenzie Jackson: Yeah. And like, and I guess that is, what's your opinion of like who's it benefiting more? Are you like from what you're saying, are you saying that like it's giving us the ability to find more things to do out, like it's giving us an edge or do you still think that like, you know, that that's kind of in response to the edge that it's kind of giving the attackers?
Norman Menz: Yeah. I mean, it's always going to be an arms race really against the clock. And, you know, the longer that organizations benefit from the fact that they have an information advantage because they know what's sensitive to them and to their organization. And they've been working for as long as they've been around to try to protect that. They have a slight time advantage to that as well. But they also know at the same time, we've got the attackers working against that using AI and they're trying to close that time window down. And so, you know, it's hard to predict who's going to prevail. I'd like to think that the good guys are going to prevail because that's what we work hard towards. But I think we're going to continuously, we're going to continuously see attacks coming that are using AI. I think what's really interesting as we look at any of them today, though, we've seen some creativity coming from, let's call them AI generated attacks, but a lot of them are still an exploit of known techniques. We haven't seen a lot of novel techniques. What you have is you have, you know, where we, where we have human actors, right? They're humans. We go to sleep, we go on vacation, et cetera. We get bored, we stop working on it. You know, we have these agents that'll sit. They're persistent. They're going to continue to run. And more likely than not, they're going to go find the vulnerabilities and the other things that have been out there. They could have been detected by people, but it's going to just happen at a greater velocity.
Mackenzie Jackson: Yeah. And I, I could actually share an example where we, where we, we saw a lot of benefits. So, you know, I think we joked a little bit. We, we have some common interest in the fact that Flare's been a user of your platform at Aikido, and you know, our developers use it in their daily workflow and things of that nature. And at the same time, traditionally as you'd expect, we do manual pen tests on our platform on a pretty regular basis. And you know, as like any organization, when you find something, you analyze it, you, you, you fix it, et cetera, triage it and whatnot. We had a really exciting. So if you talk about from a defenders use case in terms of that level of persistence, we started using your automated pen testing in the Aikido platform. And given all of the other work that we did, the one thing that was unique about this is you have access to the codebase. You can analyze. I don't want to get into the details of it. You can analyze various parts of it and then actually do the testing against the platform. And we saw really interesting results that we frankly believe that we probably wouldn't have seen or wouldn't have seen as quickly with manual pen testing.
Norman Menz: That's. It's such a cool one. I love, I love that analogy. And one of the correct things about the pen testing thing is that loop of like test, verifying it, you know, because then everything you know in this report is kind of like, you know, it's not like, oh, maybe you have an injection vulnerability here. It's like, no, we're, we're, we've gone there. I, I'm, I'm generally pretty optimistic about AI and how it's happening, but it is exhausting. Like, you know, like because you just you have to keep running until, you know, eventually it will, we will settle on on something. But crikey, right now and then, you know, your adversaries are doing crazy things and then Anthropic will do something and like, do I still have a job tomorrow? You know, like things like that. How do you generally feel about AI in terms of kind of, I know we, we just talked about like the benefits and that, but like, are you excited in this revolution? Are you concerned? You know, what's what are your kind of general thoughts on it?
Norman Menz: I'm excited. I mean, I've spent the last two decades in this space. I think this is the most exciting period that we have available for us. It's going to come with the challenges, the challenges, threats, if you will, that we've talked about here. People are going to continue to be people. They're going to be motivated by the same things that motivate them. And organizations are going to defend themselves against the same types of threats and attacks. They will. They're just going to use different things to do that, right? I think everyone knows that you can no longer have systems with vulnerabilities on them. So if for as many decades that we've been trying to close down that vulnerability gap as humans and the different process we have, we've not done a terrific job at it. I think we're going to see organizations have to adopt automated technologies to deal with this en masse because the exploits are coming very quickly. So from that part of it, I think the optimistic capability is to fully leverage automation to do that, agents, et cetera, continuing to build different layers of defense around the technology. You know, these things are really exciting. And I think that as the thing that motivates us as cybersecurity professionals are going to be the things that continue to motivate. We're just going to have a different set of tools to do that.
Mackenzie Jackson: I want to briefly interrupt this episode to thank our sponsor of the podcast, Aikido Security. Aikido Security is your complete code to cloud security solution. It secures everything from the first line of code that you write all the way through your supply chain, your dependencies. And of course, Aikido is also a leader in AI pen testing and AI code auditing. So make sure you check them out today. You can start for free at Aikido.dev. All right, enough about that. Let's get back to the episode. One of the things I'm curious about is kind of you also work in identity in other areas and we have these new types of identities appearing up in AI agents everywhere. So we've talked about LLMs and how we're able to, you know, triage staff and analyze things and that's all good. But we also gave agents tools, well, these AI things, tools. And the identity problem is so fascinating to me because we've had non-human identities for a while, but now we have non-human identities, that kind of thing for themselves that are secured, like identities that don't. And I'm just curious what your take on this new identity problem is.
Norman Menz: I think you're absolutely right here. We have, we've always had an identity problem. We talked about the identity being the new perimeter, and I think it's an acknowledgement that regardless of the controls that we put in place, as you say, we're going to have agents. Those agents are going to have a mission or mandate, right? They're going to have a set of tools and data that's available to them and they have an amazing capability to work at speed and oftentimes doing some things that we don't know. And so I think what you're going to see is we're just compounding the problem that we've seen traditionally around identity and specifically around agent identity and therefore non-human. I mean, if you look at a traditional enterprise, between human identity and non-human identity. And I'm talking about the classical sense of non-human identity, service accounts and things of that nature. Typically it was, it was seen as non-human identity to human identity was a ratio of three to one or five to one, something like that. But you got an idea of it. Now, what do we have is we probably that ratio of non-human identity to human identity is going to be even higher than that, probably easily two times that, if not more. But what do they have? I mean, service accounts classically were able to communicate from one thing to another or access to an API or whatever it happened to be. But now we have agents that have multiple identities that they have access to and they're going to be doing things that people never predicted. And so the need to understand what an agent is doing, what its configuration is, what it can have access to, what identities it has access to is going to be even more important for the organization today. And if those identities do get out, organizations need to be able to track that and remediate that immediately as well.
Mackenzie Jackson: Yeah, it is a big, it is such a big challenge now because I mean, we've talked about leaked credentials and other things as well. But just the the idea for me too of like these agents that can have access to environments, that have access to credentials. It's like the idea of leaked credential is like might not even just be a credential that's not where it's supposed to be. It's just that we just have credentials everywhere in environments of agents have access to them. They can, they can do crazy things. Is it a solvable problem or like how do I, like I'm very because I've been thinking about this too. I had a talk at B-Sides that was called the Synthetic Insiders, and it was kind of around this and I gave no solution. I just, I just talked about problems. But like what you think this is solvable? How it like controlling, understanding agents, their identities and controlling them.
Norman Menz: You know, it's funny you say that. I was on a panel a few weeks back. The title of the panel is "When AI Agents Go Shopping." So very similar to what you're talking about. And I left it feeling a little bit depressed, right? Like where we didn't give an answer. And I think the reality is the answer is there because of the same type of problems we've classically been dealing with in security. It's, it's compounded now because it's much larger. But there's going to be, we have the technology today, various, you know, we spent the last 30 plus years working on authentication, right? It's still a tricky thing. It's going to continue to be a tricky thing. This is going to make it harder. But I am optimistic, you know, that we can leverage the technology that we have. There's going to be, you know, we have gating of credentials and different systems to do that. Some organizations implement this really well. They have really good maturity. Even those organizations though, still have identity problems. And then you have the organizations that unfortunately don't maybe have the same resource or the same level of maturity where they're not vaulting credentials, you know, where you have, you know, I think best practice today is probably the agents don't have direct access to credentials, right? They're going through some type of a proxy or some type of a vault and those credentials that, so the agent never sees their credentials, right? But I've not seen a lot of organizations implementing that today.
Mackenzie Jackson: I had a conversation about this with someone today where they asked me a question that was basically, do you think we'll get on top of the problem of like leaked secrets or secrets or leaked credentials? And I was kind of thinking, well, not unless something fundamentally changes because we've always had, we've had for a long time technology to solve it. We've had dynamic secrets, we've had vaults. We've had, you know, we have everything we need to not have this problem and we have barely made a dent in the problem. So therefore, like that it won't change unless something fundamental. And then I was thinking maybe agent is that fundamental thing that shifts everything so much and becomes such a problem that we actually need to solve multiple problems at the same time which could end up being a net positive if not slightly headachy one.
Norman Menz: If you want my take on it like, and it's interesting, right? Obviously we're here at Black Hat, right? And the show's just across and everyone knows whether it was RSA, Black Hat, Def Con, like in the technology space in cyber, everyone's talking about agentic this, agentic that, SOC-er, et cetera. One of the things that's really interesting and there could be a hundred new agentic security companies in the next six months. It's probably will be the case. The question is there's the market can't sustain a hundred different ways to solve the problem or several hundred different ways to solve the problem. And one of the things I'm not hearing the industry talk about, and this gets back to the use of existing tools is building a set of standards and protocols such that these new systems and ways in which for agents to be communicate, to communicate, to authenticate, to be valid, can talk to each other in a way in a trusted model, right? And so if I'm an agent and I'm going out to another agent asking it to do something, I need a protocol or something that's bi-directional where it's accepting it. So then we can kind of, we can close the loop from kind of A-Z, if you will. And like when I look at that problem, I don't think we need to invent new technology to do that. But I think as an industry, we have to understand the fact that we've got to come together and kind of build some unanimous consent on terms. And I know it's really difficult, it's impossible, right? But there needs to be a better set of standards and protocols if we're really going to solve that because we just can't do with another agentic thing.
Mackenzie Jackson: Yeah. Can we just build something? Yeah, we've got a booth here and put AI on it. Yeah, yeah. Look, we have to finish up. We're going to try and keep the podcast to some of the sides, but we always like to play a game. It's part I'm looking forward to. Now I'm stressed because I have to try and come up with new ones for you. But I'm going to ask some classic ones because I think it's curious. I would say if you're going to build a new product today, I mean you can have a team, but you can only build the product in two ways. One is total vibe coding, but you don't understand the code, right? So it's kind of like the lovable type vibe coding. No, sure, they're lovable, but you don't know what's happening in the background. Or you have to write it character by character. Which would you rather?
Norman Menz: OK. So this will not be popular, perhaps with some of the members of my engineering team, but I'm going to go with option A. And I object to the use of the word vibe coded, but kind of given that, like, I think that's where things are going, right? And I think that, you know, I've always, it's actually funny. So I'm, I'm sorry, I'm going to dive into this one a little bit. We work in a field that has constantly improved through abstraction, right? If we look and we say, well, vibe coded apps or LLM generated or agent generated code is no good. Like look at every progression of programming languages where they had, they've always been an additional level of abstraction for the benefit of readability, you know, in some instances performance, et cetera. I see this is just another level of abstraction and it's an advancement in terms of making it easier, allowing us to be more productive, et cetera. So I'm definitely taking option A.
Mackenzie Jackson: Option A. OK, we'll do another one there. We talked about leaked credentials, so I have to ask this one here. Which would you rather if you had all your secrets API key stored in a private Git directory, or you had an application with absolutely no firewalls, no defenses, so you can't block traffic here, but you have to pick one. Which would you rather?
Norman Menz: That's a tough one. I think I would probably. So with my background kind of coming from more like classical, like defensive blue team said, I've got to go with option A again. Yeah, yeah.
Mackenzie Jackson: All right, I have to give you a hard one then. All right, we've had a lot of supply chain attacks recently. So application you can pin all your dependencies to the current version or you auto update them. So one option has like CVEs coming in which are being exploited very rapidly and one has the option of like absolute chaos and malware. And and. Which would you rather?
Norman Menz: OK, well I'm going to go with option A again. Like I, I should have mixed. Let me explain this one. So if I was talking about availability, I would actually go with option B because I think the progress like, say, I'm OK if things break, right? But the problem is what is the impact of that thing when it breaks? And that's why option B scares me, right? So option A, if I wanted something that was like from a security perspective that was secure, hardened, if you will, you know, option A is obviously the clear one. If I want to advance and move quickly and there is no malicious code in that, then I would take option B, right, because I would adopt that progress and whatnot. So yeah.
Mackenzie Jackson: I think if we got on top of some of this malware stuff, which I think we are making progress on, then the kind of auto pulling because that was like not auto pulling, but the consensus was for a long time to patch as fast as you can, right and that.
Norman Menz: I'm in the automatic patch camp, right? Then when you, when you add the other risk, right, of malware in there then and it's like, OK, we're going to take it. And there have been some very devastating supply chain attacks, right? Yeah, yeah, for sure.
Mackenzie Jackson: I'll do one more. I used the "end the three" but we'll do one more because I'm kind of interested in this one.
Norman Menz: Am I going to pick B before you go?
Mackenzie Jackson: I don't know. What would you rather to have the same password used for everything, OK, but it can be complex or, you know, whatever. Or you can use different passwords, but they're all stored in notepad on your computer.
Norman Menz: Thank goodness none of these would. I'd rather answer a different question. I would, I would, I would, I would choose option A again. And my hesitation is it is only option A, right? You're just going to test AAAI? I mean it's that you obviously, you know there's a Sophie's Choice question, but it's a good one. Thank you.
Mackenzie Jackson: Well, Norman, thanks so much for coming on the show. It's been brilliant. I hope to have you back in the future. And yeah, I can't wait to check out what you guys keep doing now over Flare.
Norman Menz: Well, thank you for the opportunity here. It's been really great.
Mackenzie Jackson: Awesome. Thanks so much.
