Latest / Expert Insights Podcast / #83. Vibe Coding: A More Secure Future?
Transcript
- 0:03SPEAKER_03: Thank you.
- 0:12Everybody is talking about the risks of VIBE coding and the
- 0:15tools designed to
- 0:15accelerate software development from AI coding assistance to
- 0:20automated copilots and
- 0:21SPEAKER_02: no-code platforms that create more security problems than they solve.
- 0:25SPEAKER_02: But what if VIBE coding could actually lead to better security outcomes?
- 0:30SPEAKER_02: I'm Joel Witts, host of the Expert Insights podcast.
- 0:33SPEAKER_02: Today,
- 0:33SPEAKER_02: I'm joined by Manoj Nair,
- 0:35the Chief Innovation Officer at Snyk to explore the changing
- 0:38face of application
- 0:39SPEAKER_02: security in the AI era.
- 0:41SPEAKER_02: From his early career building data leakage prevention at RSA,
- 0:45SPEAKER_02: through to leading cloud startups and now driving innovation at Snyk,
- 0:48Minosh has been at the front lines of some of cybersecurity's
- 0:51biggest shifts.
- 0:52SPEAKER_02: In this conversation,
- 0:53SPEAKER_02: we talk about why traditional security models are failing,
- 0:56SPEAKER_02: how shift left is evolving in the age of AI,
- 0:59and why vibe coding can either be a major liability or with the
- 1:03right guardrails
- 1:04SPEAKER_02: actually produce software that's more secure than human code.
- 1:13I've been in cybersecurity,
- 1:14I think,
- 1:14for many years,
- 1:15maybe at least 20 years now.
- 1:17My start was a data classification product that we spun out of
- 1:21a company called
- 1:21SPEAKER_01: EMC,
- 1:22SPEAKER_01: went into RSA,
- 1:23SPEAKER_01: and this was before RSA got acquired.
- 1:25SPEAKER_01: We became the team that,
- 1:27and data classification was a little bit of a problem looking
- 1:30for a solution.
- 1:31But in security,
- 1:32I had a great problem to solve in data leakage back in the day.
- 1:35SPEAKER_01: So Gen 0,
- 1:36SPEAKER_01: Gen 1 of data leakage prevention,
- 1:38SPEAKER_01: and then ended up being there five years,
- 1:41building out a portfolio of capabilities all the way from data
- 1:45leakage to Archer,
- 1:46SPEAKER_01: I don't know if you know Archer,
- 1:47SPEAKER_01: GRC,
- 1:48SPEAKER_01: governance,
- 1:48SPEAKER_01: risk and compliance.
- 1:50SPEAKER_01: We were the
- 1:51SPEAKER_01: second biggest SIM provider to Envision,
- 1:56SPEAKER_01: which was ArcSight and Envision back in the day,
- 1:58SPEAKER_01: and security intelligence,
- 1:59SPEAKER_01: event management.
- 2:00So we had this whole security management,
- 2:02security analytics portfolio being built.
- 2:05SPEAKER_01: And the last acquisition was a company called NetFitness.
- 2:08SPEAKER_01: And right around the time RSA went through a nation state breach.
- 2:11SPEAKER_01: It was the famous Aurora breach, Google and RSA.
- 2:14SPEAKER_03: Yeah.
- 2:14So it's like my portfolio of products and I'm getting to
- 2:17understand really the
- 2:18difference between security and compliance and how different
- 2:22real security attacks
- 2:24SPEAKER_01: and APTs were really unknown back then.
- 2:27SPEAKER_01: And so,
- 2:27SPEAKER_01: you know,
- 2:28SPEAKER_01: when I spent the next few years,
- 2:29SPEAKER_01: the business that was attacked was the identity business.
- 2:33After six months or so,
- 2:34I went and ran the identity business and it just...
- 2:37SPEAKER_01: Massive footprint worldwide.
- 2:38SPEAKER_01: We did 10% of anti-fraud detection and identity verification with tokens,
- 2:46SPEAKER_01: governance,
- 2:46SPEAKER_01: all that stuff.
- 2:48SPEAKER_01: Took that to the cloud, became a cloud guy suddenly.
- 2:51SPEAKER_01: And then I did my startup in the cloud space, sold it.
- 2:55SPEAKER_01: After going through my earn-out period, I was thinking what to do next.
- 2:59SPEAKER_01: So Sneak appeals to me, and the reason I joined Sneak is...
- 3:02it's so different than everything else that the security
- 3:05industry really preaches
- 3:07security even walking around blackhead i feel like it's just
- 3:11new acronyms but it's
- 3:12the same approaches like what we did back in the day 13 14
- 3:16years ago when i was at
- 3:17rsa it's like detection and response because humans think like
- 3:21that like tall you
- 3:22know walls around castles big moats maybe if he needs to put
- 3:25some spice inside to
- 3:26SPEAKER_01: find the insiders
- 3:27SPEAKER_01: This is how most of the security industry is.
- 3:29Snyk takes this approach where,
- 3:31look,
- 3:31almost every company is a software company.
- 3:33SPEAKER_01: Everyone's building software.
- 3:34And so the crown jewels are that custom software that the
- 3:37companies are building,
- 3:38SPEAKER_01: not everything else.
- 3:39SPEAKER_01: And those developers are not trained in security.
- 3:42SPEAKER_01: So how do you give
- 3:44you know,
- 3:44context automation and learning right at the root of the code
- 3:50being produced?
- 3:51SPEAKER_01: And does that result in much more trusted software?
- 3:54SPEAKER_01: And so I just found it like very appealing to me as an engineer,
- 3:58somebody who's gone through massive security issues and have
- 4:00built what I would
- 4:01consider was traditional security that fails most of the times
- 4:05when it's most
- 4:06SPEAKER_01: important to
- 4:07SPEAKER_01: a much more proactive fixing culture.
- 4:12SPEAKER_01: Like I was talking to a customer this morning,
- 4:13SPEAKER_01: the thing I love the most when you talk to successful sneak customers is,
- 4:17SPEAKER_01: this is a Fortune 10 who recently became a customer and they go,
- 4:21SPEAKER_01: 30,000 issues just disappeared.
- 4:23SPEAKER_01: We didn't tell the devs to do anything.
- 4:25SPEAKER_01: We just installed you guys and
- 4:27SPEAKER_01: They're just fixing it themselves.
- 4:28And now we're fixing it with AI,
- 4:30you know,
- 4:31so they don't even have that toil.
- 4:32SPEAKER_01: So that's a little bit of, you know, why for me, Sneak.
- 4:36SPEAKER_01: And I've been here three years now.
- 4:38And I think we're,
- 4:39we have doubled,
- 4:40more than double the size of the company.
- 4:43SPEAKER_01: Would you say sneak's really been one of the pioneers of that?
- 4:45SPEAKER_01: Is it shift left mentality?
- 4:47SPEAKER_01: Yeah.
- 4:47SPEAKER_01: So I think shift left.
- 4:48SPEAKER_01: And again,
- 4:49SPEAKER_01: it's a difficult thing because sometimes,
- 4:51SPEAKER_01: you know,
- 4:51SPEAKER_01: from a customer perspective,
- 4:53SPEAKER_01: they all start sounding like buzzwords.
- 4:54SPEAKER_01: Everyone's saying the same thing,
- 4:56SPEAKER_01: but shifting left is really shifting this from a reactive catch later,
- 5:00litter centric mindset of I'm going to put a gate and you got
- 5:03to pass that,
- 5:04SPEAKER_01: which slows down productivity to
- 5:07move the context,
- 5:08move the learning,
- 5:09move the automation right at the source.
- 5:11SPEAKER_01: So shift lab,
- 5:12SPEAKER_01: dev security,
- 5:13SPEAKER_01: DevSecOps,
- 5:15SPEAKER_01: all these are the things that I think,
- 5:17SPEAKER_01: you know,
- 5:17SPEAKER_01: Snyk's been very much a real accelerator,
- 5:20SPEAKER_01: either in creating that or making that mission real for many customers.
- 5:25SPEAKER_02: Yeah.
- 5:25SPEAKER_02: And so in terms of the actual product,
- 5:27SPEAKER_02: it is for the engineers,
- 5:29SPEAKER_02: it sits in the code,
- 5:30SPEAKER_02: it tells them when there's vulnerabilities,
- 5:31SPEAKER_02: it helps them fix them as they appear.
- 5:34SPEAKER_01: Think about the engineers or the users,
- 5:37but the security teams now don't have to become gatekeepers and
- 5:41do the testing.
- 5:42SPEAKER_01: They can do governance and policy and they get visibility.
- 5:46SPEAKER_03: Mm-hmm.
- 5:46SPEAKER_01: So how do you know that issues are being prevented?
- 5:48SPEAKER_03: Yeah.
- 5:49SPEAKER_01: But you need to have that analysis.
- 5:51SPEAKER_01: And then what is a priority?
- 5:52SPEAKER_01: Let's say you have a lot of backlog.
- 5:53SPEAKER_01: Well, let's go fix this whole class of vulnerability.
- 5:56SPEAKER_01: Now, how do you get more proactive in there?
- 5:58SPEAKER_01: So just enable.
- 6:00And usually the ratios that we see in our customers are really
- 6:03bad between security
- 6:05SPEAKER_01: engineering and AppSec and all those roles to actual devs.
- 6:08SPEAKER_01: Mm-hmm.
- 6:09SPEAKER_01: So they have to do a lot,
- 6:11SPEAKER_01: but sometimes it's,
- 6:12SPEAKER_01: people say 100 to 1,
- 6:13SPEAKER_01: it's much more than 100 to 1 is what we see sometimes.
- 6:16SPEAKER_01: There's no way that they can be the doers.
- 6:18SPEAKER_01: They have to be enablers and they have to be policy and guardrails.
- 6:22SPEAKER_01: And so that's again a culture shift that we see.
- 6:24SPEAKER_01: Definitely.
- 6:26SPEAKER_01: And while we're on the company,
- 6:27I know there's a lot of startups that have sort of appeared in
- 6:31the AppSec space,
- 6:32SPEAKER_02: being DAST or SPM or,
- 6:35SPEAKER_00: you know,
- 6:35SPEAKER_02: there's a huge ecosystem there.
- 6:37SPEAKER_02: What makes Snyk the best in that broad space?
- 6:41I think it's just,
- 6:43we don't even think of ourselves as an AppSec company.
- 6:45SPEAKER_01: So we think about it really,
- 6:47SPEAKER_01: dev security is the way we thought about it,
- 6:49SPEAKER_01: and now increasingly AI security.
- 6:51SPEAKER_01: It's AppSec, I feel like it's very traditional.
- 6:54SPEAKER_01: I'm going to do something reactively to the app.
- 6:57SPEAKER_01: And whereas dev security is like,
- 6:59SPEAKER_01: oh,
- 6:59SPEAKER_01: I'm going to enable the devs to build better security,
- 7:01SPEAKER_01: right?
- 7:01SPEAKER_01: And so that whole philosophy is a huge difference.
- 7:05SPEAKER_01: Now, there's a bunch of companies that say, oh, sure, we do that too.
- 7:08SPEAKER_01: And so now what happens is this whole approach of we're comprehensive.
- 7:12SPEAKER_01: We have the best data.
- 7:14SPEAKER_01: We're extremely fast and we're built in.
- 7:17So we're not only built in at this point enabling devs,
- 7:20we're enabling the AI devs.
- 7:22So you think about the co-pilots and the cursors and the
- 7:25windsurf of the world and
- 7:26how do you take that same context and give it to them directly
- 7:29before they show the
- 7:30SPEAKER_01: code to the dev.
- 7:31SPEAKER_02: Interesting.
- 7:32And that's something we launched at Black Hat this week is
- 7:34secure at inception.
- 7:35SPEAKER_01: Power is like this
- 7:38SPEAKER_01: full visibility of the app and produce context real time,
- 7:42SPEAKER_01: super fast,
- 7:42SPEAKER_01: like 100 times faster than any of the things that are there,
- 7:46SPEAKER_01: but much more efficient.
- 7:47SPEAKER_01: Now, I can give the context to a human.
- 7:49SPEAKER_01: The human might choose to ignore it, but the machine doesn't.
- 7:52SPEAKER_01: So this is this whole notion of context engineering.
- 7:54That's what we have been,
- 7:55like it's taken us many years to get that right.
- 7:58SPEAKER_01: So we were built into Google Gemini Codesys.
- 8:00SPEAKER_01: We can now work inside Copilot, Cursor.
- 8:03SPEAKER_01: I don't know if you
- 8:04I've heard about some of these tools,
- 8:06like we are,
- 8:07you know,
- 8:07the wipe coding generation.
- 8:09SPEAKER_01: So then with cursor, you just say, go create a code.
- 8:13We're able to create a policy file that'll make that code
- 8:16always scan by Snyk.
- 8:17SPEAKER_01: And then the results are...
- 8:19fed into Cursor,
- 8:19Cursor can feed that to the LLM,
- 8:21you get much more secure code.
- 8:23SPEAKER_01: Everyone knows,
- 8:24SPEAKER_01: I think,
- 8:24SPEAKER_01: I shouldn't say everyone,
- 8:25people now recognize that AI-generated code is more insecure
- 8:29than human-generated
- 8:30SPEAKER_01: code.
- 8:30SPEAKER_01: All kinds of studies, I think Georgetown University did one with...
- 8:35SPEAKER_01: It's like 48% of the code is insecure.
- 8:37SPEAKER_01: My favorite is a site called Baxbench, B-A-Xbench.com.
- 8:42It's a collaboration between ETH and Zurich and Berkeley and a
- 8:46bunch of researchers.
- 8:47SPEAKER_01: They benchmark back-end code, which is real enterprise code.
- 8:51Correctness and security of the latest models like a Sonnet 4
- 8:55are still like,
- 8:56SPEAKER_01: you know,
- 8:56SPEAKER_01: 60% at best.
- 8:59SPEAKER_01: So this approach is very important,
- 9:03SPEAKER_01: but it flips that from being less secure to more secure than humans.
- 9:07SPEAKER_01: Right.
- 9:07SPEAKER_01: And you supply our security contacts directly to the machine.
- 9:10SPEAKER_01: That is a whole new front.
- 9:13We all launched our AI trust platform that shows the
- 9:16comprehensiveness of the
- 9:17SPEAKER_01: platform,
- 9:18SPEAKER_01: the kinds of scanning,
- 9:19SPEAKER_01: open source code,
- 9:20SPEAKER_01: first body code,
- 9:22SPEAKER_01: infrastructure's code,
- 9:23SPEAKER_01: containers,
- 9:24SPEAKER_01: secrets,
- 9:25SPEAKER_01: this whole dynamic testing,
- 9:27SPEAKER_01: not traditional DAST,
- 9:29SPEAKER_01: API and dynamic testing for the AI era.
- 9:32SPEAKER_01: So it's not late stage, it's testing it in the pipeline.
- 9:36We're working on correlating that static analysis with dynamic
- 9:40testing results.
- 9:40SPEAKER_01: So it's almost like proactive,
- 9:43SPEAKER_01: but before deployment,
- 9:44SPEAKER_01: just have a pen tester sit right there,
- 9:46SPEAKER_01: but it's a machine doing the pen testing and it's saying,
- 9:49SPEAKER_01: look,
- 9:49SPEAKER_01: I found this issue.
- 9:50SPEAKER_01: I can exploit this issue.
- 9:51SPEAKER_01: This is a high priority issue.
- 9:53SPEAKER_01: And then think about being able to remediate that automatically.
- 9:56SPEAKER_01: That's how far we have come from where we started.
- 10:02You're listening to the Expert Insights Podcast,
- 10:04home of leading cybersecurity conversations.
- 10:07Subscribe now to get insights from top cybersecurity leaders
- 10:11straight to your inbox.
- 10:14SPEAKER_02: If you've got someone vibe coding that doesn't know how to code,
- 10:18SPEAKER_02: you get this recommendation,
- 10:19SPEAKER_02: hey,
- 10:19SPEAKER_02: this code's insecure.
- 10:20SPEAKER_02: It's not just that they ignore it.
- 10:21SPEAKER_02: They don't know what to do with it.
- 10:23SPEAKER_02: It's got to be baked in.
- 10:24SPEAKER_02: Exactly.
- 10:25SPEAKER_01: That's an even better point.
- 10:27We're seeing a lot of citizen developers and tools like Lovable
- 10:30and Replit.
- 10:31You saw that interesting story about the Saster CEO vibe coding
- 10:36as a way to delete the database.
- 10:38SPEAKER_01: Yeah.
- 10:38SPEAKER_01: I mean, these require security and safety guardrails, right?
- 10:42SPEAKER_01: Or this cursor dev who lost half a million in crypto coins because of a
- 10:47SPEAKER_01: vulnerability in a plugin.
- 10:48SPEAKER_01: There's real supply chain risks.
- 10:51And in the AI era,
- 10:52now there are newer risks with all of the AI components.
- 10:55SPEAKER_02: Yeah, a hundred percent.
- 10:57So I wanted to ask about this because obviously Sneak,
- 10:59I think launched in 2015,
- 11:00right?
- 11:00SPEAKER_02: So 10 years on and coding has completely changed, hasn't it?
- 11:05SPEAKER_02: In the enterprise space,
- 11:08are you seeing those vulnerabilities from Vibe coding actually
- 11:10putting enterprises
- 11:12SPEAKER_02: at risk?
- 11:12I mean,
- 11:12you've already touched on a couple of examples,
- 11:14but is this a serious problem?
- 11:15SPEAKER_01: It's a very serious problem.
- 11:17SPEAKER_01: We're also seeing that the bad guys are using AI to exploit.
- 11:20SPEAKER_01: So AI is being used in offensive ways.
- 11:23SPEAKER_01: We have customers,
- 11:24SPEAKER_01: for example,
- 11:24who create cohorts of devs who have access to some kind of a
- 11:29co-pilot and those who
- 11:30SPEAKER_01: don't.
- 11:30SPEAKER_01: And they'll come and share with us,
- 11:32SPEAKER_01: like,
- 11:32SPEAKER_01: look,
- 11:33these guys in nine months produced a lot more vulnerabilities
- 11:36than the guys who
- 11:37SPEAKER_01: didn't have it.
- 11:38SPEAKER_01: So I think this is where I go.
- 11:39SPEAKER_01: The awareness is much better and the risks are more real.
- 11:43SPEAKER_01: So
- 11:43The leading companies we work with don't allow an AI coding
- 11:47assistant without sneak guardrail.
- 11:49It's like sneak in the IDE,
- 11:51sneak now can enforce in the PR before the code's ever merged
- 11:54in.
- 11:54SPEAKER_01: Because once it's merged in, it's too late.
- 11:57SPEAKER_01: And so that is also very unique in what we do.
- 12:00Almost everyone who does static analysis does that after the
- 12:04build is done.
- 12:04SPEAKER_01: Yeah, yeah.
- 12:05and so ai is forcing if you thought shift left was optional
- 12:10it's not optional
- 12:11anymore and it's even more left is what we have to go than what
- 12:15we had before it's
- 12:16a great quote add a point there so in terms of the vibe coding
- 12:20yeah you mentioned
- 12:21there about how you know some teams were a lot more efficient
- 12:24with it um but
- 12:26SPEAKER_02: creating a lot more vulnerabilities in terms of the you know
- 12:29SPEAKER_02: internal conversation about Vibe coding,
- 12:30SPEAKER_02: is there a tension between security teams saying,
- 12:33SPEAKER_02: hey,
- 12:33SPEAKER_02: actually,
- 12:33SPEAKER_02: we want to stop you Vibe coding,
- 12:35SPEAKER_02: teams maybe not using Snyk,
- 12:36because you're generating these vulnerabilities versus the
- 12:39business case of maybe
- 12:40SPEAKER_02: saying,
- 12:41SPEAKER_02: well,
- 12:41SPEAKER_02: all our competitors are generating all these apps.
- 12:44SPEAKER_02: Is there a tension there?
- 12:44SPEAKER_02: How do security teams balance that speed with security that...
- 12:48SPEAKER_01: AI tools can operate.
- 12:49SPEAKER_01: Well, some people are just doing the ostrich methodology.
- 12:52So bury the head in the sand and just think that the devs are
- 12:56not using any coding agents.
- 12:58SPEAKER_01: And you'll be surprised.
- 12:59SPEAKER_01: Cursor is amazing growth.
- 13:01What very rarely do you hear an enterprise who's actually
- 13:03licensed cursor.
- 13:04SPEAKER_01: So who's using it?
- 13:06SPEAKER_01: It's the devs.
- 13:06SPEAKER_01: They're paying 20 bucks a month or whatever because it's making them...
- 13:10SPEAKER_01: And so there's shadow AI in the web coding tool chain.
- 13:14SPEAKER_01: There's definitely tension.
- 13:16SPEAKER_01: But in the end,
- 13:17what we generally see is very rarely do I find a security team
- 13:21who has successfully
- 13:22SPEAKER_01: blocked the adoption.
- 13:23SPEAKER_01: No CEO wants to be left behind on...
- 13:27the AI race and every board is asking about what are you doing
- 13:30with AI for productivity.
- 13:31SPEAKER_01: It doesn't matter how laggard an organization you are,
- 13:34SPEAKER_01: people are asking the question,
- 13:35SPEAKER_01: how are you using AI?
- 13:36SPEAKER_01: So I think it's rare,
- 13:39SPEAKER_01: but do you see the occasional,
- 13:41SPEAKER_01: like,
- 13:41SPEAKER_01: I'm just going to assume they're not using it.
- 13:44SPEAKER_01: So I find different than the cloud era,
- 13:47SPEAKER_01: that in the AI era,
- 13:49more security teams want to proactively figure out how to
- 13:52enable safe adoption of
- 13:54SPEAKER_01: AI.
- 13:54SPEAKER_01: Say no, or block it, because they're just not going to succeed.
- 13:58And so the pressure to harvest the productivity and innovation
- 14:02is immense.
- 14:03SPEAKER_01: So safe adoption...
- 14:05SPEAKER_01: and understanding what is the best way to have productivity.
- 14:07SPEAKER_01: Ideally, security is invisible.
- 14:09SPEAKER_01: And that's what we're enabling with things like secured innovation.
- 14:12SPEAKER_01: The other thing that they need is,
- 14:14SPEAKER_01: so there's white coding,
- 14:15SPEAKER_01: but then there's also the next era of AI native apps.
- 14:19SPEAKER_01: So,
- 14:20SPEAKER_01: you know,
- 14:20SPEAKER_01: back in the day,
- 14:21SPEAKER_01: I think,
- 14:21SPEAKER_01: was it Marc Andreessen who said software will lead the world?
- 14:24SPEAKER_01: We're now seeing AI is going to eat software.
- 14:26SPEAKER_01: Yeah.
- 14:26So software,
- 14:27as you know,
- 14:28doesn't exist because AI native apps are so different.
- 14:31SPEAKER_01: And those AI native apps are agents, models, and they're fully agentic.
- 14:37SPEAKER_01: They're agent to agent interactions.
- 14:39And this is now the next,
- 14:40you know,
- 14:41the next mountain to climb in the sense from security teams.
- 14:44SPEAKER_01: They have to understand what the teams are doing.
- 14:47SPEAKER_01: What agents are they using?
- 14:48SPEAKER_01: What models are they using?
- 14:50SPEAKER_01: Are they using DeepSeq?
- 14:52SPEAKER_01: Are they building MCP servers that could be a backend?
- 14:54Because somebody joked about this this week,
- 14:57actually,
- 14:58the S in MCP stands for security.
- 15:01SPEAKER_01: So this is highly prolific protocol,
- 15:04SPEAKER_01: but people are not,
- 15:05SPEAKER_01: there's no authorization built in initially.
- 15:08SPEAKER_01: Now there's authorization.
- 15:09SPEAKER_01: People don't
- 15:10SPEAKER_01: You know, people are finding new risks all the time.
- 15:13SPEAKER_01: So we acquired a company.
- 15:14SPEAKER_01: It's a spin-out from ETH Research called Invariant Labs.
- 15:19SPEAKER_01: These are the kinds of researchers who are finding these new risks.
- 15:23SPEAKER_01: Part of what we need to do is educate the market on the risks.
- 15:26SPEAKER_01: They're the ones who found the GitHub MCP server exploit,
- 15:29which got a lot of people thinking about if Microsoft's GitHub
- 15:33MCP server is
- 15:34SPEAKER_01: exploitable,
- 15:35SPEAKER_01: then imagine...
- 15:37Joe speeds a shop MCP server that's downloaded by my dev from
- 15:43cursor.directory
- 15:44SPEAKER_01: that's not even owned by cursor.
- 15:46It's like this crowdsourced place that people are listing MCPs
- 15:50for any dev to download and
- 15:53SPEAKER_01: It's like WhatsApp MCP and this,
- 15:55SPEAKER_01: and like,
- 15:55why would you want a WhatsApp connected to your cursor to wipe
- 15:58code and share the
- 15:59SPEAKER_01: code with your,
- 16:00SPEAKER_01: like,
- 16:00SPEAKER_01: so there's all kinds of crazy use cases.
- 16:03SPEAKER_01: And so understanding that and then finding the security of that.
- 16:06SPEAKER_01: So one of the things we launched this week was this,
- 16:09toxic flow analysis detection capability to find issues in MCP
- 16:14like the one that
- 16:15SPEAKER_01: caused the GitHub server,
- 16:17SPEAKER_01: MCP server exploit.
- 16:18SPEAKER_01: And so that's the other set of things we're doing.
- 16:21We're turning on visibility for all security teams on what AI
- 16:25components are
- 16:27SPEAKER_01: being used and built in your code, like the EU AI Act.
- 16:31SPEAKER_01: I mean, you have to have visibility.
- 16:33SPEAKER_01: You need to know, you cannot say, I don't know what was AI used.
- 16:36SPEAKER_01: You need to have guardrails.
- 16:38SPEAKER_01: You need to have secure code generation.
- 16:40SPEAKER_01: So these are all things that,
- 16:43SPEAKER_01: you know,
- 16:43we are now having solutions in the market or researching and
- 16:47bringing out new
- 16:47SPEAKER_01: solutions.
- 16:48SPEAKER_02: Yeah, 100%.
- 16:50SPEAKER_02: Are you generally optimistic or pessimistic about the use of AI?
- 16:53SPEAKER_02: I mean,
- 16:54SPEAKER_02: obviously,
- 16:55SPEAKER_02: there are some,
- 16:56SPEAKER_02: I guess,
- 16:56SPEAKER_02: security teams that,
- 16:57SPEAKER_02: as you say,
- 16:58SPEAKER_02: either looking the other way or kind of,
- 17:00SPEAKER_02: you know,
- 17:00SPEAKER_02: concerned about how AI is going to,
- 17:03SPEAKER_02: you know,
- 17:03SPEAKER_02: I mean,
- 17:03SPEAKER_02: you've looked at a long term future here around how AI gets software.
- 17:08Do you think that's going to be a net positive for security or
- 17:11just different challenges?
- 17:12SPEAKER_01: I think both are true.
- 17:13SPEAKER_01: I'm,
- 17:13SPEAKER_01: you know,
- 17:15SPEAKER_01: 100% glass half full here that first of all,
- 17:17SPEAKER_01: there's nothing that history of humanity shows that if anything improves
- 17:20SPEAKER_01: productivity,
- 17:21SPEAKER_01: we're going to adopt it.
- 17:22SPEAKER_01: So now the question is, how can you do it in this era?
- 17:24This is the fastest any technology has touched the breadth of
- 17:30humanity that,
- 17:31SPEAKER_01: you know,
- 17:32SPEAKER_01: any,
- 17:32SPEAKER_01: like think about a billion plus chat GPT users now.
- 17:35SPEAKER_01: at the pace at which that has happened.
- 17:37SPEAKER_01: So yes, you're going to use it, enable safety and done right.
- 17:41SPEAKER_01: We're using AI to secure AI, all kinds of different kinds of AI.
- 17:46But you know,
- 17:46that you couldn't do what I said,
- 17:48like secure at inception or anything.
- 17:50If I was doing static rules-based checking of code,
- 17:53that would be slow,
- 17:54that would be too late.
- 17:55SPEAKER_01: And so just thinking of that philosophy and just being curious about it.
- 18:00SPEAKER_01: And some of our most advanced customers themselves are using AI to as
- 18:03SPEAKER_01: you know, really be much more proactive in defending themselves.
- 18:08SPEAKER_01: So, you know, yeah, call me optimistic.
- 18:10Yeah,
- 18:10definitely don't bury your head in the stand and think it's not
- 18:13happening.
- 18:13SPEAKER_01: It's happening with or without you.
- 18:15SPEAKER_01: Fantastic.
- 18:17Is there anything that we haven't covered that you wanted to
- 18:19kind of talk about?
- 18:20SPEAKER_01: I think,
- 18:21SPEAKER_01: you know,
- 18:21SPEAKER_01: part of,
- 18:23SPEAKER_01: we touched on it,
- 18:24such a huge potential when you see this explosive growth of
- 18:27Vibe coding to truly
- 18:29SPEAKER_01: secure code better than humans would.
- 18:32SPEAKER_01: That's the thing that people need to get educated on.
- 18:34So,
- 18:34you know,
- 18:35for your readership,
- 18:35that'll be like a,
- 18:37it's almost like this,
- 18:38oh,
- 18:38wow.
- 18:38SPEAKER_01: This is not the negative.
- 18:39SPEAKER_01: This is a net positive.
- 18:41SPEAKER_01: Right.
- 18:42SPEAKER_01: But how, right?
- 18:43SPEAKER_01: How do you do that?
- 18:44SPEAKER_01: And I think the other thing is just,
- 18:45SPEAKER_01: you know,
- 18:46SPEAKER_01: people are building AI apps,
- 18:47SPEAKER_01: whether you know it or not.
- 18:48SPEAKER_03: Mm-hmm.
- 18:49and so understanding that driving visibility driving awareness
- 18:54you know proper like
- 18:55let's not do in this era what we did in every other technology
- 18:59era like the cloud
- 19:00like after the fact i'm going to put some gatekeepers to catch
- 19:03issues yeah it
- 19:04doesn't matter how beautiful the graph is and how great the
- 19:07interface is it's too
- 19:09SPEAKER_01: late so like really thinking about how to be proactive or
- 19:13SPEAKER_01: as AI software, let that AI be secure by default.
- 19:17SPEAKER_01: Yeah.
- 19:17SPEAKER_01: I think those are the two big things.
- 19:19SPEAKER_01: Like secure code at inception, secure the stack by default.
- 19:23SPEAKER_02: Yeah.
- 19:23So the big takeaway on bibe coding then is if it's insecure
- 19:27code,
- 19:27it's an issue.
- 19:28SPEAKER_02: But if that code has the right guardrails,
- 19:31SPEAKER_02: the right security behind it,
- 19:33SPEAKER_02: then it can be more secure than human code.
- 19:36SPEAKER_02: Really interesting.
- 19:36SPEAKER_02: Yeah.
- 19:37SPEAKER_02: It's a great perspective.
- 19:38SPEAKER_02: Thank you very much.
- 19:39SPEAKER_02: I really appreciate it.
- 19:42Bye.
- 19:55This podcast was produced by Expert Insights,
- 19:58home of leading cybersecurity conversations.
- 20:01Subscribe now to get insights from top cybersecurity leaders
- 20:04straight into your inbox.
- 20:06Why not check out our weekly decrypted cybersecurity briefings
- 20:09and our Game
- 20:10SPEAKER_00: Changers conversations with innovative individuals.