Latest / Expert Insights Podcast / #70 - RSAC Sandbox Special: CalypsoAI and ProjectDiscovery
Transcript
- 0:04SPEAKER_01: Hello,
- 0:04SPEAKER_01: and welcome to the Expert Insights Podcast,
- 0:07SPEAKER_01: your go-to place for in-depth cybersecurity conversations.
- 0:13SPEAKER_01: This episode is a little different to usual.
- 0:16SPEAKER_01: It's a double bill.
- 0:17The RSAC Sandbox event hosts 10 different startups,
- 0:21all vying for attention.
- 0:22SPEAKER_01: Today, we've got interviews with two of these companies.
- 0:25SPEAKER_01: We speak to Project Discovery's Andy Chow.
- 0:28SPEAKER_01: But first, we have an interview with Calypso AI.
- 0:31Expert Insights Head of Content,
- 0:32Joel Witts,
- 0:33speaks to Calypso AI's Donifer Casey.
- 0:36The interview focuses on the challenges and innovations in
- 0:39securing Gen AI applications.
- 0:41SPEAKER_01: The pair talk about balancing security with innovation,
- 0:44SPEAKER_01: the role of red teaming in identifying vulnerabilities,
- 0:47SPEAKER_01: and the various use cases for AI adoption in enterprises.
- 0:52We're kind of a strange company in that we're a US company,
- 0:55US founded,
- 0:56US domiciled.
- 0:58You know,
- 0:58almost 90% of the company at this point is in Dublin and
- 1:01Ireland.
- 1:01SPEAKER_02: All the R&D is in Dublin, which is great.
- 1:05SPEAKER_02: you know, we'll probably have to rebalance a bit to the U.S.
- 1:09It was kind of,
- 1:11it was a smart move,
- 1:13I think,
- 1:13to build the R&D team in Dublin.
- 1:15SPEAKER_02: It's cheaper.
- 1:16SPEAKER_02: We get access to really good talent.
- 1:19SPEAKER_02: I think probably more of a stable talent pool.
- 1:21SPEAKER_02: We've got a really strong tenure for our talent in R&D and kind of in the
- 1:26SPEAKER_02: engineering and data science.
- 1:28SPEAKER_02: And then we have access to a kind of a
- 1:30a European pool and a Dublin pool and the cool thing like great
- 1:34talent in Dublin
- 1:35but they all work for big multinationals that landed in Ireland
- 1:39as a means of going
- 1:41SPEAKER_02: kind of taking taking on Europe or a lot of the innovation or the kind of
- 1:48playbooks are already built and you're just kind of rewriting
- 1:50them for Europe or
- 1:51SPEAKER_02: for whatever marketer in the UK or France or whatever.
- 1:54SPEAKER_02: So people,
- 1:54when they get the opportunity to kind of write their own
- 1:56playbook from scratch,
- 1:58SPEAKER_02: there's a certain profile of people that just jump on that.
- 2:01And so,
- 2:02like,
- 2:03obviously Google are over 10,000 in Dublin now,
- 2:06Facebook,
- 2:07you know,
- 2:086,000,
- 2:087,000.
- 2:08I'm hoping more and more for Ireland that what we see is a lot
- 2:11more startups
- 2:11spinning out of those big companies and we start to innovate
- 2:14for ourselves and have
- 2:15SPEAKER_02: our own companies.
- 2:15SPEAKER_02: Because even Stripe, which would be kind of the
- 2:18the darling in Ireland at the moment those guys the Collison's
- 2:20came to the US to
- 2:22find that company and kind of reversed it back into Ireland
- 2:24eventually you know
- 2:25right interesting but DC it'd be great to talk about your
- 2:28background and how you
- 2:29got into Calypso and into the security space sure yeah so
- 2:32before Calypso I spent
- 2:34almost a decade growing a company called Qualtrics from Series
- 2:40B when I joined it
- 2:41SPEAKER_02: up to
- 2:43SPEAKER_02: A series of exits, actually.
- 2:46SPEAKER_02: First is acquisition by SAP for eight and a half billion dollars.
- 2:50SPEAKER_02: 2018, 2019, we IPO'd at SAP.
- 2:53SPEAKER_02: At one point,
- 2:53I think it might have been worth about $45 billion if you
- 2:55blinked at the stock
- 2:57SPEAKER_02: market.
- 2:58SPEAKER_02: But IPO'd about 25, 30 billion, which is a great ride to be on.
- 3:03SPEAKER_02: And then ultimately, we're brought private again by Silver Lake.
- 3:07SPEAKER_02: Another 12 and a half of this.
- 3:08SPEAKER_02: So three exits for that leadership team, quite a lot.
- 3:10SPEAKER_02: And a lot learned along the way.
- 3:13SPEAKER_02: A lot of exciting things happened.
- 3:14SPEAKER_02: One interesting thing in the middle of all of that as well is COVID,
- 3:16which kind of changed a lot of things for me because I was
- 3:19originally supposed to
- 3:19SPEAKER_02: move.
- 3:19SPEAKER_02: I've got four kids and a wife, right?
- 3:21SPEAKER_02: And we were supposed to move to Utah with Qualtrics.
- 3:24SPEAKER_02: And then COVID came and I had all my stuff shipped to the US.
- 3:27SPEAKER_02: And I had my family kind of looking.
- 3:29SPEAKER_02: We're picking out locations for schools and all this kind of stuff.
- 3:31SPEAKER_02: And then we couldn't go.
- 3:32SPEAKER_02: And we rented out our house in Dublin as well to somebody.
- 3:35SPEAKER_02: We hadn't moved out of it yet.
- 3:37SPEAKER_02: So we managed to stay in it and kind of...
- 3:39shutters came down on the world and ended up doing the job
- 3:42remote like lots of us
- 3:43and then continued on as the kind of only executive based
- 3:46outside the US and lived
- 3:48on a plane for a good few years and we IPO'd during that time
- 3:51so you know the usual
- 3:53SPEAKER_02: go ring the bell in NASDAQ moment that everybody gets.
- 3:56SPEAKER_02: Well,
- 3:56I got that at home in my kitchen where I was asked to send in a
- 4:00picture with a
- 4:00SPEAKER_02: blank background and I superimposed the leadership team on a podium,
- 4:03SPEAKER_02: which is disappointing.
- 4:04But I guess that,
- 4:05you know,
- 4:05everybody has their disappointments through that time.
- 4:07SPEAKER_02: So that was super interesting for me.
- 4:09SPEAKER_02: But one of the most impactful things was through that time,
- 4:12SPEAKER_02: I met a guy called Jimmy White,
- 4:13SPEAKER_02: right?
- 4:13SPEAKER_02: And he worked with me in the Qualtrics European business early on.
- 4:17SPEAKER_02: And then he left the Qualtrics business five and a half years ago.
- 4:20And he came and joined Calypso as kind of the founding engineer
- 4:24and founding CTO
- 4:25and built the product from scratch and stayed in contact with
- 4:28him over that time.
- 4:29Called me one day and asked me about,
- 4:30you know,
- 4:31hey,
- 4:31do you want to go for breakfast?
- 4:32And if you know Jimmy,
- 4:33it's quite rare that he offers to pay for anybody's breakfast.
- 4:35SPEAKER_02: So I jumped it to chance.
- 4:37SPEAKER_02: And of course, he wanted to talk about kind of
- 4:39SPEAKER_02: this new phase in the company's history.
- 4:42SPEAKER_02: We're kind of shifting to kind of looking at the Gen AI world,
- 4:46SPEAKER_02: going from the space room,
- 4:47SPEAKER_02: which is the government and federal space,
- 4:49into the commercial world and going from securing of the
- 4:52building of ML-driven
- 4:53SPEAKER_02: models to securing kind of interest in the ML world as well.
- 4:57SPEAKER_02: So a lot of changes and I saw a lot of areas that could have impact.
- 5:01loved working with Jimmy at Qualtrics so jumped the fence and
- 5:04joined the company
- 5:05officially in March of last year and that's kind of never we'd
- 5:08never look back it's
- 5:09only just over a year but it feels like I think seven years
- 5:12we're living we're all
- 5:13living in dog years at the moment in the AI world and there's
- 5:17so much change in so
- 5:18many different innovations every day every week it certainly
- 5:21keeps you interested
- 5:22SPEAKER_02: right so that's kind of
- 5:23SPEAKER_02: what's brought me to the space, really.
- 5:26SPEAKER_03: So the tagline, agentic warfare is here.
- 5:28SPEAKER_03: Yeah.
- 5:29SPEAKER_03: So it'd be great to talk about the platform,
- 5:30SPEAKER_03: what you guys do and what challenge you're here to solve.
- 5:33SPEAKER_02: Absolutely.
- 5:33So the big challenge we're really here to solve is we see a
- 5:38huge opportunity for
- 5:39society with Gen.AI and huge opportunities for enterprises and
- 5:43business within
- 5:44SPEAKER_02: that.
- 5:44And the first barrier that businesses and enterprises face when
- 5:47they're adopting Gen.AI is
- 5:50SPEAKER_02: can I do it,
- 5:50SPEAKER_02: but can I do it in a secure and safe way that's not going to impact my
- 5:55SPEAKER_02: shareholders,
- 5:56SPEAKER_02: my customers,
- 5:57SPEAKER_02: my employees,
- 5:58SPEAKER_02: my partners,
- 5:59SPEAKER_02: whoever it is,
- 6:00SPEAKER_02: right?
- 6:00And so we saw an opportunity to come in and be a critical part
- 6:03of that conversation.
- 6:05And so we started off last year helping our customers build
- 6:10their own applications
- 6:11SPEAKER_02: that sit on Gen.AI.
- 6:13SPEAKER_02: And of course,
- 6:14SPEAKER_02: applications,
- 6:14SPEAKER_02: if they're to do anything kind of useful for business,
- 6:17SPEAKER_02: and do anything sophisticated,
- 6:19SPEAKER_02: We'll probably have data that's maybe sensitive.
- 6:23SPEAKER_02: We'll have applications that are kind of nuanced in how they operate.
- 6:27And so we saw the opportunity to come in and help sit between
- 6:30the applications
- 6:31SPEAKER_02: those enterprises are building and the LLMs that they're using.
- 6:36SPEAKER_02: We sit between,
- 6:37SPEAKER_02: as our platform sits between our customers,
- 6:39our enterprise customers and LLMs and their customers and do a
- 6:44few things in a
- 6:45SPEAKER_02: defensive posture.
- 6:47First is,
- 6:48you know,
- 6:49AI is a new surface through which a company can be attacked,
- 6:53right?
- 6:54And so we will help protect companies against new kind of
- 6:56threat vectors that exist
- 6:58SPEAKER_02: now and are evolving all the time at a very rapid pace.
- 7:01SPEAKER_02: probably a more rapid pace than we've ever seen before.
- 7:04SPEAKER_02: So that's the first thing.
- 7:05SPEAKER_02: The second thing is a DLP use case, right?
- 7:07SPEAKER_02: There's a lot of sensitive data flowing to and from these applications.
- 7:11SPEAKER_02: And that's only going to be further increased with agents.
- 7:13SPEAKER_02: We can talk about that in a second, right?
- 7:14But we want to make sure that no PAI or IP or anything else
- 7:18escapes out of the
- 7:19SPEAKER_02: enterprise in a way that would be damaging to anybody.
- 7:22SPEAKER_02: And so we help with that.
- 7:23And then the third one,
- 7:24which is super novel for Gen AI,
- 7:25is Gen AI is non-deterministic.
- 7:28SPEAKER_02: The beauty and the power of it is that it'll be,
- 7:31SPEAKER_02: it's almost like conversing with a person,
- 7:33SPEAKER_02: but you never quite can control what any given individual can say.
- 7:36SPEAKER_02: And it's the same with Gen.AI.
- 7:38SPEAKER_02: So we have some certain customers that say,
- 7:40SPEAKER_02: hey,
- 7:40SPEAKER_02: like we have a food ingredients customer,
- 7:42SPEAKER_02: as an example,
- 7:43who once wanted to build an application that represents or that
- 7:48suggests recipes
- 7:49SPEAKER_02: using the ingredients of that company to its customers.
- 7:52SPEAKER_02: But they wanted to make sure that,
- 7:54SPEAKER_02: hey,
- 7:54don't suggest ingredients from another manufacturer or don't
- 7:59allow from a brand
- 8:00SPEAKER_02: perspective,
- 8:01SPEAKER_02: you know,
- 8:01SPEAKER_02: recipes for cocktails or things with a lot of alcohol or whatever.
- 8:05SPEAKER_02: Or by the way,
- 8:05SPEAKER_02: our company has not always done the right thing for our own kind of,
- 8:09SPEAKER_02: for society.
- 8:10SPEAKER_02: Probably don't mention that either because we don't want a brand risk.
- 8:13SPEAKER_02: And so they're not traditional cyber use cases,
- 8:15SPEAKER_02: but they're still,
- 8:15SPEAKER_02: you know,
- 8:16kind of an integral part of how we help customers to do kind of
- 8:18things safely,
- 8:19SPEAKER_02: securely as kind of a broader package.
- 8:21So there's a defensive posture that we have now that we sit in
- 8:25place and protect
- 8:26SPEAKER_02: those applications being built,
- 8:27SPEAKER_02: right?
- 8:28SPEAKER_02: Just last quarter, we...
- 8:30SPEAKER_02: recognized that all that support and help was post-production.
- 8:34There's a great opportunity for us to actually come
- 8:35pre-production to help
- 8:36SPEAKER_02: customers think about things and get ahead of some of those risks early.
- 8:40SPEAKER_02: And so we launched a red teaming product.
- 8:42SPEAKER_02: You mentioned Agentec Warfare.
- 8:44SPEAKER_02: And the red teaming product allows us to go and help customers with,
- 8:47SPEAKER_02: after they selected a use case,
- 8:49SPEAKER_02: what model should I pick that's the best for my use case?
- 8:52SPEAKER_00: Right.
- 8:52Models are,
- 8:53in themselves,
- 8:53there's been a massive focus on model performance.
- 8:56SPEAKER_02: But up until maybe DeepSeq,
- 8:58SPEAKER_02: a few months back,
- 8:59SPEAKER_02: people hadn't really been talking about,
- 9:00SPEAKER_02: hey,
- 9:00SPEAKER_02: how secure are these models?
- 9:01Or am I happy to use a model because I might be leaking my data
- 9:05to some third party
- 9:06SPEAKER_02: or,
- 9:06SPEAKER_02: you know,
- 9:07SPEAKER_02: might not actually be the right model for me.
- 9:08So at least DeepSeq did us a favor there and kind of bringing
- 9:11that to the fore.
- 9:12SPEAKER_02: So we help our customers actually pick the right model.
- 9:15SPEAKER_02: We've got a leaderboard.
- 9:17We have three different types of attacks that we use on models
- 9:20to identify how
- 9:21SPEAKER_02: vulnerable they are to a broad range of things.
- 9:24SPEAKER_02: We've got a leaderboard that we brought out.
- 9:25SPEAKER_02: We've got what we call our signature attacks.
- 9:29SPEAKER_02: We develop 10,000 new signature attacks a month.
- 9:32we're up to i think 50 000 new attacks now um and those 50 000
- 9:36attacks every month
- 9:37when we develop the new attacks we point them at the latest and
- 9:40greatest models the
- 9:41essentially the how those models stack up against those attacks
- 9:45we will rank the
- 9:46models we'll give them a score and then ranking and the score
- 9:48is a bit more
- 9:49SPEAKER_02: sophisticated than what you see else
- 9:52SPEAKER_02: out there on the market to date.
- 9:53SPEAKER_02: We had to develop our own score.
- 9:54SPEAKER_02: Previously it was,
- 9:55SPEAKER_02: you know,
- 9:55SPEAKER_02: it was 400 attacks and,
- 9:59SPEAKER_02: you know,
- 9:59SPEAKER_02: what percentage of those attacks got through.
- 10:01SPEAKER_02: But that's not really a true reflection of kind of,
- 10:05of the damage that can be done or lack thereof because one
- 10:09attack might be just
- 10:10trying to convince a model to push your granny down the stairs
- 10:12and while nobody
- 10:13wants to push your granny down the stairs it's not going to
- 10:15necessarily be
- 10:17materially damaging to a business whereas another attack might
- 10:19actually drive your
- 10:20model to actually hand over very sensitive data and that can be
- 10:24catastrophic for
- 10:24business so that has more material impact and those two should
- 10:27not be weighted the
- 10:28SPEAKER_02: same.
- 10:29So we have an inbuilt kind of formula to identify how impactful
- 10:34those attacks will
- 10:35SPEAKER_02: be on a business that will be weighted in the ranking.
- 10:38And so we have a CASI score that stack ranks all of the models
- 10:42against each other.
- 10:43No surprise,
- 10:43the last couple of months,
- 10:45Anthropic has been coming out on top.
- 10:47SPEAKER_02: One interesting thing is the most recent Anthropic model,
- 10:49SPEAKER_02: Claude37,
- 10:50SPEAKER_02: it's actually less secure than its predecessor.
- 10:53SPEAKER_02: And that's by design, right?
- 10:54SPEAKER_02: Because, yeah, exactly.
- 10:55SPEAKER_02: So there's a compromise you want to make.
- 10:58SPEAKER_02: being super conservative and super secure,
- 11:00you probably then start kind of dialing your security controls
- 11:03too tightly.
- 11:04SPEAKER_02: And then customers probably say,
- 11:05SPEAKER_02: well,
- 11:05SPEAKER_02: hang on,
- 11:05SPEAKER_02: it's not as usable as something that is maybe slightly less secure,
- 11:08SPEAKER_02: but,
- 11:09SPEAKER_02: you know,
- 11:09SPEAKER_02: kind of opens up the,
- 11:10SPEAKER_02: you know,
- 11:11SPEAKER_02: lets some of the performance come through as well.
- 11:14SPEAKER_02: That's going to be a challenge, right?
- 11:15SPEAKER_02: And that's a balance we go through with all our customers, right?
- 11:18SPEAKER_02: Our customers come to us and say, hey, I want to block everything.
- 11:23SPEAKER_02: And we say, well, that's great.
- 11:24SPEAKER_02: I understand why you want to say that, but we can help you do that.
- 11:27SPEAKER_02: In fact, there's many open source tools out there to do that.
- 11:29SPEAKER_02: But like,
- 11:30SPEAKER_02: if you're,
- 11:32SPEAKER_02: use the analogy,
- 11:32if you're a bouncer in a nightclub and you stop everybody from
- 11:35coming in that
- 11:35SPEAKER_02: night,
- 11:37Good news,
- 11:37there'll be no trouble in the nightclub,
- 11:38but you won't have any customers.
- 11:39SPEAKER_02: And it's the same with securing LLMs.
- 11:42If you're willing to block everything,
- 11:44then you actually won't be able to innovate.
- 11:45And what we help our customers do is to dial up or dial down
- 11:49your posture based on
- 11:50SPEAKER_02: the needs of your business,
- 11:51SPEAKER_02: based on your use case,
- 11:52SPEAKER_02: based on applications and users to make sure that you can innovate,
- 11:57SPEAKER_02: but be secure.
- 11:57SPEAKER_02: And there's always a balance.
- 11:59And sometimes we can create a window in which you can actually
- 12:02facilitate a
- 12:02SPEAKER_02: conversation with an LLM or a customer that
- 12:05allows you to unlock,
- 12:07to continue to innovate,
- 12:07but not take all the risks.
- 12:23just to be clear on the use cases do you cover all because you
- 12:27know obviously
- 12:27you've got there's so many different ways that companies are
- 12:29using ai whether
- 12:30that's to their customers with chat bots or that sort of stuff
- 12:33or internally with
- 12:34knowledge sharing yeah um yes we cover everything yeah is the
- 12:38is the answer and by
- 12:39the way i'll i'll get back to the red team in a second but yes
- 12:42we cover everything
- 12:42right what we see is there's kind of a maturity curve that
- 12:45customers go through
- 12:46SPEAKER_02: right the first thing often if you come to a customer super mature
- 12:49They're either saying,
- 12:50hey,
- 12:51I built this application in the corner of the company.
- 12:53SPEAKER_02: It's not in production and it's for an internal use case.
- 12:56SPEAKER_02: Can you help us get it live?
- 12:57SPEAKER_02: And yeah, of course we can, but there's usually not a high risk there.
- 13:00Maybe they want their employees to access ChatGPT or some other
- 13:05kind of chat
- 13:06interface where they just use GenAI in the way that we all
- 13:09started using GenAI a
- 13:10SPEAKER_02: while back.
- 13:11SPEAKER_02: right?
- 13:11SPEAKER_02: That's a very kind of standardized way of using Gen AI,
- 13:15SPEAKER_02: but it's still an unlock for many people.
- 13:17Where people start getting more advanced is applications that
- 13:20they're building on top.
- 13:21SPEAKER_02: And they might start with internal face applications,
- 13:23SPEAKER_02: again,
- 13:24SPEAKER_02: so lower risk because you're kind of keeping it in the family.
- 13:26SPEAKER_02: But you might do something in HR that maybe is a CV screener.
- 13:30You put your resume through the screener and it picks out the
- 13:33top five candidates for
- 13:34a job where you might have had 200 different applications and
- 13:37then it ranks them
- 13:38and whatever else right super useful low risk though but like
- 13:41good there is
- 13:42probably a little bit of uh risk of a person's steadfast you
- 13:46know whatever escaping
- 13:48yeah but internal use case probably lower risk but we still
- 13:50help customers with
- 13:51that and then you you start saying okay maybe higher stakes
- 13:54internal use case
- 13:55finance organization wants to use it to analyze internal
- 13:57numbers with sensitive
- 13:59numbers but still internal so we can help there and then
- 14:01businesses get more and
- 14:02more brave flip it externally low-end customer facing use cases
- 14:06standard kind of
- 14:06support requests that kind of thing again more complex because
- 14:09it's external facing
- 14:10and more risk basic and then and on and on and on you go up the
- 14:14stack until you're
- 14:15you're basically having very sophisticated complex
- 14:18conversations your customers
- 14:19with potentially sensitive data in the mix as well if it got
- 14:24into the wrong hands
- 14:25so there's uh you know there's a big spectrum there we help our
- 14:28customers wherever
- 14:29they are in a maturity curve and because we work a lot across a
- 14:33broad range of
- 14:34customers often we will have seen what works and what doesn't
- 14:37before them so some
- 14:38of the value we have in conversations with our customers is of
- 14:41course our software
- 14:42SPEAKER_02: but also how to deploy it to get the
- 14:44SPEAKER_02: best bang for their buck.
- 14:45SPEAKER_02: And then back to the red teaming,
- 14:46SPEAKER_02: just to not overtalk that,
- 14:48SPEAKER_02: we got the tens of thousands of attacks that we throw at people,
- 14:50SPEAKER_02: which are the signature attacks.
- 14:51We have operational attacks,
- 14:52which kind of attack the infrastructure behind the models.
- 14:55So there's new types of attacks that exist there,
- 14:58like a DDoS attack in the old world.
- 15:01SPEAKER_02: There's software that's been around a long time to stop that.
- 15:04SPEAKER_02: But a similar DDoS type of attack in the JDI world would be,
- 15:08SPEAKER_02: hey,
- 15:08SPEAKER_02: I'm just going to prompt the engineer here,
- 15:09SPEAKER_02: give me the history of the world of 40 million pages.
- 15:12SPEAKER_02: that's going to sail through any existing security protections you have.
- 15:15The model is going to receive that request and effectively shut
- 15:18down because you'll
- 15:19have given it such a workload that it will be inoperable for
- 15:22any other business use
- 15:24SPEAKER_02: cases for the time being.
- 15:25You hit it with two or three of those and any business critical
- 15:29functions you have
- 15:30SPEAKER_02: will be effectively inoperable.
- 15:32SPEAKER_02: And so we help with those as well.
- 15:34SPEAKER_02: And then the third one, which is the exciting one, the agentic warfare.
- 15:37SPEAKER_02: We have trained our own agent
- 15:39SPEAKER_02: to go with all the latest attack techniques,
- 15:42SPEAKER_02: all the history of our experience attacking previous models,
- 15:45SPEAKER_02: apps,
- 15:46SPEAKER_02: and agents.
- 15:46SPEAKER_02: And we give that agent a custom intent.
- 15:50SPEAKER_02: So it's like, hey, go get me.
- 15:52SPEAKER_02: There's a new app we're launching.
- 15:53SPEAKER_02: It's customer-facing.
- 15:54Go get me Joel's social security number and bank details or
- 15:57whatever,
- 15:58right?
- 15:58And it will go and find a way to autonomously compromise that
- 16:02app and come back
- 16:02SPEAKER_02: with a way to kind of get those details.
- 16:04SPEAKER_02: And it will come, hey, I was able to get those details.
- 16:06SPEAKER_02: It took me an hour and a half.
- 16:07SPEAKER_02: Here's how I did it.
- 16:08SPEAKER_02: Here's the prompts that I use.
- 16:10SPEAKER_02: Here's the kind of chain of thought.
- 16:11SPEAKER_02: And there's lots of different techniques coming out all the time.
- 16:14SPEAKER_02: It knows all of them, right?
- 16:15And it goes to find,
- 16:16yeah,
- 16:16and it's very,
- 16:17very effective at finding gaps in the fences.
- 16:19The reason it's so effective and a company of our size can do
- 16:22that is because
- 16:23SPEAKER_02: behind that,
- 16:24SPEAKER_02: we've got agents doing our own
- 16:26research behind the scenes they're they're processing hundreds
- 16:31of different papers
- 16:32on new and novel attacks each week behind the scenes um we've
- 16:36got other agents that
- 16:37work with those agents say okay that's good new intel now let's
- 16:40let's develop novel
- 16:41SPEAKER_02: attacks on the back of that
- 16:43And we've got other agents then that kind of remember what
- 16:45attacks work and what
- 16:46SPEAKER_02: don't and what sequences they work in against which models.
- 16:49SPEAKER_02: And so all of those agents are working for us as part of our team,
- 16:53SPEAKER_02: to scale a lot quicker,
- 16:54SPEAKER_02: but also allow our red teaming products to be a lot more effective.
- 16:58SPEAKER_02: In particular, when you look at not just other
- 17:01SPEAKER_02: technologies that are kind of just using 400,
- 17:04SPEAKER_02: 600,
- 17:07800 attacks that are just standard off the shelf.
- 17:07SPEAKER_02: But here are other individuals that might take six,
- 17:10SPEAKER_02: eight weeks to red team a model or an app,
- 17:13SPEAKER_02: cost you a lot of money,
- 17:14SPEAKER_02: and ultimately,
- 17:16actually not end up with the same level of results so the
- 17:18quality isn't there either,
- 17:19right?
- 17:20SPEAKER_02: So we're reducing timeframes,
- 17:21SPEAKER_02: we're increasing quality and reducing cost,
- 17:23SPEAKER_02: which is very counterintuitive to the normal.
- 17:25SPEAKER_02: Usually something's got to give there.
- 17:27SPEAKER_02: And that's something,
- 17:27the value we can bring to our customers before they even get to
- 17:29production.
- 17:30SPEAKER_02: And that's kind of the latest and greatest innovation on agentic warfare.
- 17:33SPEAKER_02: That's what we call agentic warfare,
- 17:34is using our own agents to go and expose vulnerabilities before
- 17:38the bad guys do.
- 17:39SPEAKER_02: And you can bet if we've got that technology,
- 17:41then there's people out there thinking about how to use that
- 17:43technology for
- 17:44SPEAKER_02: malicious purposes.
- 17:45thank you so much up next we have andy chow project discovery
- 17:53ceo he discusses how
- 17:55the company was founded as well as the importance of community
- 17:59in vulnerability
- 17:59management and how their platform aims to automate and simplify
- 18:03the process for
- 18:04SPEAKER_01: security teams
- 18:09SPEAKER_03: So Ani, great to meet you.
- 18:11It'd be great to start with an introduction to yourself and how
- 18:14you got into the
- 18:14SPEAKER_03: security space and the sort of journey with Project Discovery.
- 18:17SPEAKER_00: Yeah.
- 18:18SPEAKER_00: So I started in security back in 2012, 2013.
- 18:22SPEAKER_00: I came from the finance world.
- 18:25So I did investment banking and then I joined a venture capital
- 18:29firm that did a lot
- 18:30SPEAKER_00: of investments in cybersecurity.
- 18:32SPEAKER_00: And my first investment was in a company called Prolexic back in 2013.
- 18:36SPEAKER_00: It was a company that was later acquired by Akamai.
- 18:39And I remember doing diligence on this company and they were
- 18:44founded in Florida.
- 18:45SPEAKER_00: They originally started protecting like gaming websites.
- 18:48SPEAKER_00: And I just went deep into the world of hackers.
- 18:51SPEAKER_00: And this was like a DDoS mitigation company.
- 18:54SPEAKER_00: And I just thought hackers and security was the coolest thing.
- 18:59SPEAKER_00: And ever since then, it's been a huge fascination of mine.
- 19:03I made a number of investments on the venture capital side and
- 19:06then ultimately
- 19:07SPEAKER_00: decide I want to become an operator.
- 19:09And it was just such an obvious choice to join a security
- 19:12company because I was I
- 19:14SPEAKER_00: just this has been a passion of mine for many years.
- 19:16SPEAKER_00: So finally in the thick of things and I couldn't be having more fun.
- 19:20SPEAKER_03: Yeah.
- 19:20SPEAKER_03: Fantastic.
- 19:21And can you tell us a bit about Project Discovery,
- 19:23the platform that you guys do and
- 19:25SPEAKER_03: kind of the background behind founding the company?
- 19:27SPEAKER_00: Yeah,
- 19:27we are an open source powered security company on a mission to
- 19:30solve vulnerability
- 19:31SPEAKER_00: management.
- 19:32SPEAKER_00: Vulnerability management has been around for decades,
- 19:35SPEAKER_00: and so have the scanners that many people are using today to build their
- 19:39SPEAKER_00: vulnerability management program.
- 19:41And these scanners just really haven't evolved to detect the
- 19:45security risks that
- 19:46SPEAKER_00: companies today face.
- 19:47And so we built a open source tool called Nuclei a number of
- 19:52years ago that's just
- 19:54SPEAKER_00: been picked up by all the security engineers,
- 19:56SPEAKER_00: the ethical hackers,
- 19:58SPEAKER_00: everyone around this global community.
- 20:00And we're building a commercial product around it so that
- 20:02companies can replace
- 20:04SPEAKER_00: their scanners like Tenable and Qualys with our cloud solution.
- 20:08SPEAKER_00: And how does,
- 20:09SPEAKER_00: you know,
- 20:10SPEAKER_00: what's the difference,
- 20:11SPEAKER_00: what's better about your platform than the legacy ones?
- 20:13SPEAKER_00: Yeah,
- 20:14SPEAKER_00: it's all about automating what a researcher,
- 20:17a security engineer would do to verify if a vulnerability is
- 20:21actually exploitable.
- 20:22SPEAKER_00: You know,
- 20:22SPEAKER_00: 20 years ago,
- 20:23SPEAKER_00: it was definitely,
- 20:24SPEAKER_00: it was more than good to look at version numbers and say,
- 20:28SPEAKER_00: hey,
- 20:28you have an application running that's out of date and there's
- 20:31a security advisory
- 20:32SPEAKER_00: for it.
- 20:32SPEAKER_00: But today, with the tens of thousands of CVEs out there,
- 20:36SPEAKER_00: That is no longer enough.
- 20:37SPEAKER_00: And if you look at any vulnerability management dashboard,
- 20:40SPEAKER_00: you're seeing tens of thousands,
- 20:42SPEAKER_00: if not hundreds of thousands of critical vulnerabilities.
- 20:45And that's just such an overwhelming place to be in for any
- 20:48security team.
- 20:49SPEAKER_00: And we're starting with what can an attacker actually exploit?
- 20:52Let's start there and then start with really good data and then
- 20:56go from there in
- 20:57SPEAKER_00: terms of triaging and figuring out what to prioritize.
- 21:00SPEAKER_00: And how important is the community aspect of the platform?
- 21:03SPEAKER_00: It's critical.
- 21:04SPEAKER_00: You know, security started as a community effort.
- 21:08SPEAKER_00: And I think over the years we've moved away from that,
- 21:10SPEAKER_00: but I think the security industry is ready to embrace that again.
- 21:13SPEAKER_00: I think if you look at all of our most critical vulnerabilities,
- 21:19SPEAKER_00: template contributions have come from our community.
- 21:22SPEAKER_00: And it's from people all over the world.
- 21:24And it's just so great to see the community come together
- 21:27because the sooner we get
- 21:29SPEAKER_00: a detection template,
- 21:30the sooner the companies can actually run that and see where
- 21:34they're exposed or
- 21:34SPEAKER_00: vulnerable.
- 21:35SPEAKER_00: And that really is the only way to stay ahead of attackers.
- 21:38SPEAKER_00: Yeah, absolutely.
- 21:40SPEAKER_03: You mentioned that the speed of getting those vulnerabilities remediated.
- 21:44How important is that speed in the automation and kind of how
- 21:47is the platform
- 21:47SPEAKER_03: helping organizations to fix things faster?
- 21:50SPEAKER_00: Absolutely.
- 21:51SPEAKER_00: We have a great case study out with Elastic,
- 21:53SPEAKER_00: and we kind of talked about this in the presentation yesterday,
- 21:57SPEAKER_00: but Next.js was a critical vulnerability that came out last month,
- 22:02SPEAKER_00: and everyone was looking to see if they were exposed.
- 22:05And our team at Elastic was able to use our nuclei template
- 22:08that we collaborated
- 22:09with our community on and scanned their entire perimeter for
- 22:13exposed or vulnerable
- 22:14SPEAKER_00: hosts and come away with,
- 22:15SPEAKER_00: hey,
- 22:15SPEAKER_00: we're safe.
- 22:16SPEAKER_00: And they got that insight within minutes.
- 22:21SPEAKER_00: And our champion there was telling us that without Project Discovery,
- 22:26SPEAKER_00: he would have had to create an incident response ticket.
- 22:29He would have had to go through and manually triage all the
- 22:33hundreds of false
- 22:33SPEAKER_00: positives that might have come out from the traditional scanner.
- 22:36SPEAKER_00: Because all the traditional scans we're looking at is just,
- 22:38SPEAKER_00: did you have an outdated version of Next.js?
- 22:40SPEAKER_00: Right.
- 22:41So we saved dozens of hours and also just freed up a lot of the
- 22:44back and forth
- 22:45SPEAKER_00: where as soon as you start having one team in security,
- 22:47SPEAKER_00: talk to another team and creating a ticket,
- 22:49SPEAKER_00: all of a sudden things just get really complicated.
- 22:51SPEAKER_03: Yeah.
- 22:52So it's about simplifying,
- 22:53moving things faster and just improving the whole process.
- 22:55SPEAKER_00: A hundred percent.
- 22:56SPEAKER_00: Yeah.
- 22:56SPEAKER_03: Yeah.
- 22:56SPEAKER_03: Fantastic.
- 22:58And who are your customers at the moment and how do you think
- 23:00the platform will grow in scale?
- 23:01SPEAKER_00: What are your sort of ambitions?
- 23:03SPEAKER_00: I mean,
- 23:03we've got customers from Fortune 500 companies down to smaller
- 23:07startups that are
- 23:08just looking to get a view on their perimeter and check the
- 23:11compliance box.
- 23:12One of the great things about our solution is we not only help
- 23:15provide security
- 23:16SPEAKER_00: value,
- 23:16SPEAKER_00: we also check the compliance box too.
- 23:18So you can use our scan reports to pass SOC 2,
- 23:21PCI and other compliance frameworks.
- 23:24So yeah,
- 23:24we're kind of,
- 23:26we have different sized products and tiers for any size
- 23:28company.
- 23:29SPEAKER_00: And yeah,
- 23:30we love to talk to anyone who's frustrated with their scanner
- 23:33and wants more
- 23:34SPEAKER_00: actionable insights.
- 23:35SPEAKER_00: Fantastic.
- 23:36SPEAKER_03: And you guys won the Innovation Sandbox yesterday.
- 23:38SPEAKER_03: So, you know, fantastic stuff.
- 23:40SPEAKER_03: What do you think set you guys apart from the strong competition?
- 23:45SPEAKER_00: You know, I think...
- 23:46Vulnerability management is a big problem,
- 23:48and it's been a big problem for many years.
- 23:50SPEAKER_00: So it's not a new or hot thing that anyone likes to talk about,
- 23:53but I think you just sit down with any CISO and you review
- 23:56where their team is
- 23:57SPEAKER_00: spending time,
- 23:57SPEAKER_00: where their dollars are being spent,
- 23:59and so much of it is still bogged down in very paper-based
- 24:03manual workflows and
- 24:03SPEAKER_00: vulnerability management.
- 24:05SPEAKER_00: While I think that,
- 24:06SPEAKER_00: and AI is absolutely a big issue,
- 24:10SPEAKER_00: and I'm really excited to see the innovation there,
- 24:12SPEAKER_00: I think it was a testament to a long-standing,
- 24:14SPEAKER_00: old,
- 24:15SPEAKER_00: but still a very big problem.
- 24:17And I also think the problem here,
- 24:19I would say,
- 24:20is we can solve it,
- 24:21right?
- 24:22SPEAKER_00: I mean,
- 24:22it's not like these templates are rocket science and we're
- 24:25doing some magic AI
- 24:26SPEAKER_00: behind the scenes.
- 24:27We're just automating the basic things a security engineer
- 24:30would be doing anyway.
- 24:31So I think when you combine a big problem with the ease of
- 24:34solving it and then the
- 24:36SPEAKER_00: community effort,
- 24:37SPEAKER_00: I think it really is a very inspiring approach to innovation.
- 24:41SPEAKER_03: Yeah,
- 24:41I think the community in particular seemed like such a unique
- 24:43part of the pitch and
- 24:44SPEAKER_03: it's great to see fostering that collaboration.
- 24:47SPEAKER_03: What were the challenges in building that?
- 24:49Did you find that it was hard to bring people together or was
- 24:51it actually once you
- 24:52SPEAKER_03: provided the platform?
- 24:53SPEAKER_00: When we first released Nuclei, we had 40 templates.
- 24:58SPEAKER_00: Today we have 10,000, but back then it was just 40.
- 25:02SPEAKER_00: And we had no idea if this was going to get picked up or not.
- 25:05SPEAKER_00: But some of us come from the world of ethical hacking.
- 25:08And we knew that there was an opportunity to automate a lot of
- 25:13the things that
- 25:14SPEAKER_00: people were doing to find bugs that attackers could exploit.
- 25:17SPEAKER_00: So I think we believed in the mission.
- 25:19SPEAKER_00: But yeah,
- 25:19SPEAKER_00: early days was just responding to every GitHub comment,
- 25:24SPEAKER_00: just really engaging with the community.
- 25:26SPEAKER_00: But I think the community really appreciated what we built.
- 25:29SPEAKER_00: And it wasn't actually our first open source tool.
- 25:31And we're kind of like HashiCorp,
- 25:32so we have many different open source tools.
- 25:34SPEAKER_00: Nuclei is obviously the most well-known one.
- 25:37SPEAKER_00: But I think it helped some of our other open source tools,
- 25:40SPEAKER_00: helped also create a little bit of that,
- 25:42SPEAKER_00: like,
- 25:42SPEAKER_00: OK,
- 25:43SPEAKER_00: we know how to build a community here and got us going a little bit.
- 25:47SPEAKER_00: How would you like to see the company grow over the next five years?
- 25:50Well,
- 25:50I mean,
- 25:51first,
- 25:52our community comes first and we wouldn't be where we are
- 25:54without them.
- 25:55SPEAKER_00: So we're continuing to invest in our community.
- 25:57SPEAKER_00: One of the questions the judges had was,
- 25:58SPEAKER_00: you know,
- 25:59SPEAKER_00: how do you think about commercial and open source?
- 26:00SPEAKER_00: Yeah.
- 26:01And we are committed to keeping Nuclei open source and our
- 26:04community templates free and open.
- 26:06So I think we want to continue to invest in that because I
- 26:08think there's still a
- 26:09lot of people who haven't heard about Nuclei who haven't seen
- 26:11the value of it.
- 26:12SPEAKER_00: Yeah.
- 26:13SPEAKER_00: And I think the bigger the community, the more innovation we can pursue.
- 26:18SPEAKER_00: So I think that's still our number one goal.
- 26:20Second thing and why we're here is we want to grow the
- 26:22awareness that we do have a
- 26:23SPEAKER_00: commercial solution.
- 26:25SPEAKER_00: So many CISOs out there have no idea that their teams are using nuclei.
- 26:29And I think there's such a big opportunity to make it a turnkey
- 26:32solution for them.
- 26:33SPEAKER_00: So it's not like,
- 26:34SPEAKER_00: I mean,
- 26:35some of our customers have spent one or two full-time employees
- 26:37standing up an
- 26:38SPEAKER_00: entirely homegrown solution.
- 26:40And there's so many more valuable things that people can do
- 26:43than like standing up
- 26:44SPEAKER_00: their homegrown DIY nuclei scanners.
- 26:47SPEAKER_03: Yeah, absolutely.
- 26:48My final question for you is kind of what excites you most in
- 26:51the security space at the moment?
- 26:52SPEAKER_00: Wow, that's actually a hard question because I don't,
- 26:56I mean,
- 26:56this is gonna sound really generic,
- 26:59but AI is changing everything.
- 27:01SPEAKER_00: And it's even changing how we think about our solution.
- 27:04I mean,
- 27:0434 years ago,
- 27:06it was just around templates,
- 27:08but today we've trained an LLM to generate
- 27:12SPEAKER_00: hundreds, thousands of nuclei templates from public POCs.
- 27:17And where we're going with this is actually creating custom or
- 27:22unique templates for
- 27:24SPEAKER_00: security companies,
- 27:25SPEAKER_00: all of their security issues.
- 27:26So it's kind of think of that like as security as code,
- 27:29but in the form of nuclei templates.
- 27:31SPEAKER_00: And so I think AI is unlocking so much of that.
- 27:34So I think we're constantly challenging ourselves to rethink
- 27:37what are the things
- 27:38SPEAKER_00: that will be different in a few years
- 27:40SPEAKER_00: while also not creating like a completely black box,
- 27:44SPEAKER_00: like AI solution,
- 27:45SPEAKER_00: which I think we really want to avoid.
- 27:47SPEAKER_00: Like we want to empower the security engineers.
- 27:49SPEAKER_00: We want the outputs to still be nuclei templates,
- 27:52SPEAKER_00: not something that like,
- 27:54SPEAKER_00: you know,
- 27:54SPEAKER_00: you just kind of get,
- 27:55SPEAKER_00: like it's a black box and you get some results and you're like,
- 27:58SPEAKER_00: I don't know how I got that.
- 27:59SPEAKER_00: Right, right.
- 28:00SPEAKER_03: Is that really important to the practitioners that they
- 28:03have that or is it just more your philosophy if you look at dev
- 28:06and infra and why
- 28:07ai has been such a force multiplier for engineers people aren't
- 28:11getting a just a
- 28:12black box software out of it they're getting code that they can
- 28:15then go in and
- 28:16modify right and so we have to take i think there's an
- 28:19opportunity to take that
- 28:20same approach to security where i don't think security
- 28:23engineers want just
- 28:25SPEAKER_00: something at the end and like how did you get there and i can't modify it
- 28:28And that's kind of been one of the founding kind of approaches
- 28:31for Nuclei for us is
- 28:33SPEAKER_00: that visibility and transparency.
- 28:35SPEAKER_03: Yeah, absolutely.
- 28:36Do you guys see yourselves as like the next generation of
- 28:39vulnerability management
- 28:40SPEAKER_03: or are you a replacement?
- 28:41SPEAKER_03: Are you like the evolution?
- 28:42SPEAKER_00: Absolutely.
- 28:43SPEAKER_00: Yeah, yeah.
- 28:44SPEAKER_00: And I think it doesn't even just stop with vulnerability management.
- 28:47We actually have nuclei templates today that can do cloud
- 28:50misconfiguration checks.
- 28:51SPEAKER_00: Right.
- 28:51SPEAKER_00: And we're nowhere near to being like,
- 28:54SPEAKER_00: you know,
- 28:54SPEAKER_00: a CNAP or anything,
- 28:55SPEAKER_00: but a lot of the basic stuff,
- 28:57SPEAKER_00: I mean,
- 28:58you can connect nuclei into your AWS CLI and you can start
- 29:01running checks for
- 29:02SPEAKER_00: things.
- 29:02So I think this community power template driven approach to
- 29:06security can apply to a
- 29:08SPEAKER_00: lot of other different categories in security.
- 29:10SPEAKER_00: And that's what gets me really excited.
- 29:11SPEAKER_00: Yeah.
- 29:12SPEAKER_00: Fantastic.
- 29:12SPEAKER_00: Thank you so much.
- 29:22This podcast was produced by Expert Insights,
- 29:25home of leading cybersecurity conversations.
- 29:28Subscribe now to get insights from top cybersecurity leaders
- 29:31straight into your inbox.
- 29:33Why not check out our weekly decrypted cybersecurity briefings
- 29:36and our Game
- 29:37SPEAKER_01: Changers conversations with innovative individuals.