Latest / Expert Insights Podcast / #88. AI vs AI at RSAC: What 8 Security Leaders Really Think
Transcript
- 0:00SPEAKER_01: AI was supposed to make security teams more efficient,
- 0:03SPEAKER_01: more productive,
- 0:04SPEAKER_01: and ultimately more secure.
- 0:06SPEAKER_01: That was the promise.
- 0:07SPEAKER_01: But at RSAC 2026, the reality looked very different.
- 0:12Industry leaders spent the week warning that AI is introducing
- 0:15new risks faster
- 0:17SPEAKER_01: than anyone can manage them,
- 0:18SPEAKER_01: that human oversight of AI systems simply cannot keep up,
- 0:22SPEAKER_01: and that the CISO role is being stretched to breaking point.
- 0:26SPEAKER_01: At last year's RSAC, the big talking point was human in the loop.
- 0:30SPEAKER_01: This year, experts warned that humans are too slow.
- 0:34SPEAKER_01: That shift happened in a single year.
- 0:37I'm Alex Zawonetsky from Expert Insights,
- 0:40and this is our RSAC 2026 Roundup.
- 0:42SPEAKER_01: MUSIC
- 0:49Everyone in your organisation is a developer now,
- 0:52even if they don't know it.
- 0:54SPEAKER_01: Anyone can start creating landing pages,
- 0:56interactive dashboards or full business applications without
- 1:00any coding ability.
- 1:02SPEAKER_01: But when you give an AI agent a task, a lot happens behind the scenes.
- 1:06These models are downloading skills,
- 1:08installing packages and connecting to open source servers.
- 1:12SPEAKER_01: and they're doing it all without human oversight or controls.
- 1:16SPEAKER_01: Attackers know this,
- 1:17and they're already starting to exploit the trust humans are
- 1:20putting in AI models.
- 1:22Vanta's Kush Kashyap warned that shadow AI is exponentially
- 1:26bigger than shadow IT
- 1:28SPEAKER_01: ever was,
- 1:29SPEAKER_01: and the difference is important.
- 1:31SPEAKER_01: Shadow IT was people using unauthorised apps.
- 1:35SPEAKER_01: Shadow AI is agents that don't follow set courses.
- 1:38SPEAKER_01: They can go off script,
- 1:40SPEAKER_01: they're vulnerable to prompt injection,
- 1:42SPEAKER_01: and they can take action nobody asked them to.
- 1:45SPEAKER_00: Shadow AI is much more exponentially bigger than Shadow IT.
- 1:49Zscaler's Deepen Desai described a real-world incident that
- 1:53shows just how fast
- 1:55SPEAKER_01: this can go wrong.
- 1:56A developer installed an AI framework on their dev machine for
- 2:00experimentation.
- 2:01SPEAKER_01: A week later, the bot was running on the organization's CICD pipeline.
- 2:06SPEAKER_01: Nobody had moved it there.
- 2:08and they were they were clueless on how it jumped from here to
- 2:12there because nobody
- 2:13did anything but it's the agent's ability to do that next level
- 2:18thing based on
- 2:18whatever you're asking it to do which is where you know whether
- 2:22it is lateral
- 2:22propagation whether it is moving around accessing things that
- 2:26you didn't want it to
- 2:28but it's it's it has access it will access someone is tinkering
- 2:32with it the attack
- 2:33SPEAKER_03: surface again is very very different
- 2:35SPEAKER_01: And it's not just the agents themselves.
- 2:37The skills and plugins they are downloading are becoming a new
- 2:41attack vector.
- 2:42SPEAKER_01: Snyk's managed Nair compared them to malware.
- 2:45SPEAKER_03: So we did research on skills.
- 2:47We found skills are actually being exploited and there's like
- 2:50malware in there.
- 2:51SPEAKER_03: Three lines of English.
- 2:53SPEAKER_03: can compromise you and run.
- 2:54SPEAKER_03: So this is not like a binary and an executable.
- 2:57SPEAKER_03: And so that is, you know, we're having all of this adoption.
- 3:02SPEAKER_03: We're having people trust their agents more and more.
- 3:05Agents are having a lot more power and then they're being
- 3:08compromised,
- 3:08right?
- 3:09And so you now have to think about the problems we talked about
- 3:12in the past,
- 3:12SPEAKER_03: which is the code is still not very secure.
- 3:14SPEAKER_03: It's getting better.
- 3:15SPEAKER_03: Yeah.
- 3:16SPEAKER_03: The artifact itself has other kinds of issues that didn't exist before.
- 3:21SPEAKER_03: So it has more business authorization and business logic kind of issues.
- 3:26SPEAKER_03: It's got an environmental, there's a new supply chain.
- 3:30SPEAKER_03: And then the behavior is still non-deterministic.
- 3:32SPEAKER_03: So it could go off the rails.
- 3:34SPEAKER_03: And so this whole notion of great, adopt, it will happen.
- 3:39SPEAKER_03: But these risks are already there.
- 3:41SPEAKER_03: And so you got to safely adopt.
- 3:43You have to think about this agenting that life cycle not being
- 3:47something that you
- 3:48SPEAKER_03: just continue in the same way.
- 3:49SPEAKER_03: And that is the big conversation to be had.
- 3:53SPEAKER_01: Nair also warned that agents are becoming a new class of insider threat,
- 3:58SPEAKER_01: except it's not a human,
- 3:59SPEAKER_01: it's an agent that's going to be the reason you get compromised.
- 4:03SPEAKER_01: Agents can go rogue through prompt injection,
- 4:06SPEAKER_01: poison servers,
- 4:07SPEAKER_01: or just by making errors during large sessions.
- 4:10SPEAKER_01: There have already been real-world examples,
- 4:13SPEAKER_01: AWS's Kero agent deleting a database,
- 4:16SPEAKER_01: and Meta experiencing outages linked to agentic systems.
- 4:20Desai argued that the same zero trust principles we apply to
- 4:23users need to be
- 4:24SPEAKER_01: applied to agents.
- 4:26SPEAKER_01: If agents are sitting on a flat network where they can poison each other,
- 4:29SPEAKER_01: you're creating a huge new attack surface.
- 4:32Every organization,
- 4:34either it's a CEO mandate or board level mandate to...
- 4:38adopt ai so sysos are not in a place to stop that but sysos are
- 4:45definitely in a
- 4:45place to become that business enablers with this technology
- 4:48where hey how do i help
- 4:50my organization securely adopt it for security teams the
- 4:54message is clear if you
- 4:56don't know what agents are running in your environment what
- 4:58skills they're
- 4:59SPEAKER_01: installing and what access they have you could already be at risk
- 5:04SPEAKER_01: On Monday's opening CISO panel,
- 5:06Google Cloud's Francis D'Souza pointed to breakout times that
- 5:09have collapsed from
- 5:10SPEAKER_01: tens of minutes to under 30 seconds.
- 5:13SPEAKER_01: With AI agents,
- 5:14SPEAKER_01: attackers can constantly be scanning your network perimeter,
- 5:18SPEAKER_01: and a breach can happen before anyone even notices.
- 5:21SPEAKER_01: D'Souza's argument was blunt.
- 5:23The response has to be AI fighting AI,
- 5:26not humans fighting AI,
- 5:28otherwise we won't win.
- 5:30The threat landscape is shifting so rapidly that micromanaging
- 5:34AI tools is not
- 5:35SPEAKER_01: viable when organisations are running hundreds and thousands of agents.
- 5:39SPEAKER_01: By the time a human reviews an action, the moment has passed.
- 5:43SPEAKER_01: John Ramsey,
- 5:44SPEAKER_01: the VP and GM of Google Cloud Security,
- 5:46told us that he thinks we're heading to a point where the
- 5:49relationship between
- 5:50SPEAKER_01: humans and AI flips entirely.
- 5:53I think there's an evolution here and the evolution is today
- 5:57we're using agents to help humans.
- 6:01SPEAKER_03: I think the next model,
- 6:02SPEAKER_03: if we need to get to fully autonomous,
- 6:05SPEAKER_03: is that instead of the agents supporting the humans,
- 6:08SPEAKER_03: the humans will support agents.
- 6:11SPEAKER_01: Rather than humans stuck in the loop,
- 6:13SPEAKER_01: the shift is towards humans overseeing the loop,
- 6:16SPEAKER_01: setting guardrails,
- 6:17SPEAKER_01: defining policies,
- 6:18SPEAKER_01: and letting AI operate within them autonomously.
- 6:21The organisations that figure out that governance models have
- 6:25to come first will be
- 6:26SPEAKER_01: the ones that can use AI securely.
- 6:29And the reason this governance question is so urgent is because
- 6:32of what's happening
- 6:32SPEAKER_01: on the attack side.
- 6:34AI-driven phishing and social engineering is the number one
- 6:37threat that CISOs are
- 6:39SPEAKER_01: concerned about in 2026.
- 6:41That's according to Expert Insights' own CISO Confidence and
- 6:44Investment Trends report.
- 6:46SPEAKER_01: The rise of generative AI has meant an explosion in phishing campaigns,
- 6:51SPEAKER_01: which are easier than ever for hacking gangs to spam out at scale.
- 6:55SPEAKER_01: Not only is phishing more widespread,
- 6:57it's also more sophisticated and harder for email security
- 7:00controls to spot.
- 7:02SPEAKER_01: Eric Cron, CISO advisor at KnowBefore, shared a sobering statistic.
- 7:06SPEAKER_01: The average phishing campaign used to send out about 10 identical emails.
- 7:11SPEAKER_01: Today, that number has dropped to 1.8.
- 7:14SPEAKER_01: Almost every email is slightly different with different text,
- 7:17SPEAKER_01: different subjects,
- 7:18SPEAKER_01: and sometimes even different payloads,
- 7:21SPEAKER_01: all to get around email security filters.
- 7:25SPEAKER_03: And I think the last time I talked to one of our guys over there,
- 7:29they were saying it had dropped from an average campaign size
- 7:33of like 10 emails
- 7:35SPEAKER_03: being the same to 1.8.
- 7:37SPEAKER_03: Really?
- 7:37SPEAKER_03: Wow.
- 7:38SPEAKER_03: So almost every single one is different.
- 7:41SPEAKER_03: So you could have five phishing emails to the same organization.
- 7:44SPEAKER_03: They're all slightly different.
- 7:45SPEAKER_03: Right.
- 7:46SPEAKER_03: And it's not just the text or the subject.
- 7:48SPEAKER_03: It can be the payload that's in it too gets tweaked a little bit.
- 7:51SPEAKER_03: Right.
- 7:52They're doing whatever they can to get around the filters,
- 7:55around the endpoint protection.
- 7:57Ironically,
- 7:58the more polished an email looks,
- 7:59the less likely it is to be human-written.
- 8:02The old red flags,
- 8:03the typos,
- 8:04and the outlandish claims,
- 8:05well,
- 8:06they're disappearing.
- 8:07Molly McLean-Sterling,
- 8:08senior director of Proofpoint,
- 8:10explained how this is changing the game.
- 8:13It's really funny that a perfect email is now almost more of a
- 8:19tell than it was before.
- 8:23SPEAKER_03: I was in a session.
- 8:24SPEAKER_03: Gosh, I can't remember.
- 8:26SPEAKER_03: I think it was a few months ago.
- 8:27SPEAKER_03: Yeah.
- 8:29SPEAKER_03: at Educause and it was with a bunch of CISOs at major universities.
- 8:36SPEAKER_03: And one of them said,
- 8:37the way that I tell whether something is fishing now from my
- 8:41boss is whether it's
- 8:42SPEAKER_03: funny.
- 8:42SPEAKER_03: My boss is not funny.
- 8:44And so I think that gets down to the amount of context that we
- 8:51have to have.
- 8:52SPEAKER_01: And it's not just email anymore.
- 8:54SPEAKER_01: Phishing has become a multi-channel, multi-stage problem.
- 8:58SPEAKER_01: It's in your LinkedIn, your Teams, your Slack.
- 9:01Attackers are deliberately pivoting conversations to encrypted
- 9:04platforms like
- 9:05SPEAKER_01: WhatsApp because security teams can't monitor these channels.
- 9:09SPEAKER_03: So it's not a one and done phishing email anymore.
- 9:14SPEAKER_03: And I mean,
- 9:15SPEAKER_03: those still happen,
- 9:16SPEAKER_03: but it's in your LinkedIn,
- 9:18SPEAKER_03: it's in your teams,
- 9:19SPEAKER_03: it's in your Slack.
- 9:20SPEAKER_03: And you need to be able to prevent tax across that.
- 9:25SPEAKER_01: The voice cloning piece is particularly alarming.
- 9:28No Before's Eric Kron spoke to us about how quickly these lures
- 9:31can be spun up and
- 9:32SPEAKER_01: how convincing they are.
- 9:35SPEAKER_03: It doesn't take any time at all to clone the voice.
- 9:37SPEAKER_03: Right.
- 9:38SPEAKER_03: Um, I did a talk in New York and we cloned the CEO.
- 9:42SPEAKER_03: It took me 16 seconds to clone his voice perfectly.
- 9:4716 seconds of audio off a YouTube interview and having those
- 9:51tools at their
- 9:52SPEAKER_03: disposal,
- 9:52SPEAKER_03: be able to do stuff like that is crazy.
- 9:55And imagine too,
- 9:57if you get an email that says,
- 9:59I need you to wire out $150,000.
- 10:02SPEAKER_03: And you're kind of like, ah, all right.
- 10:04But then you get a phone call that says,
- 10:07I'm in the airport,
- 10:08got to get on a plane.
- 10:09SPEAKER_03: I need you to wire that money.
- 10:10SPEAKER_03: I just sent you the email.
- 10:11SPEAKER_03: Let me know when you're done.
- 10:12SPEAKER_03: Yeah.
- 10:13SPEAKER_03: Yeah.
- 10:14SPEAKER_03: Well, how affected is that going to be versus just, uh, send money?
- 10:18SPEAKER_03: Yeah.
- 10:18SPEAKER_03: Yeah.
- 10:19SPEAKER_01: But it's not just humans being the target anymore.
- 10:22SPEAKER_01: As people begin delegating tasks to AI agents,
- 10:25SPEAKER_01: from inbox management to processing invoices,
- 10:28SPEAKER_01: those agents become new targets for social engineering.
- 10:32And they may introduce risks that look a lot like traditional
- 10:35phishing,
- 10:36but at machine speed.
- 10:38Molly McLean-Sterling described a scenario where a phishing
- 10:41email is too long to
- 10:42SPEAKER_01: read,
- 10:42SPEAKER_01: so the recipient runs it through co-pilot to get a summary.
- 10:46SPEAKER_01: Hidden in the email,
- 10:47SPEAKER_01: in the white text invisible to the human eye,
- 10:50SPEAKER_01: are instructions that AI reads and executes.
- 10:53SPEAKER_01: So with the prompt injection, I mean that piece.
- 10:56SPEAKER_03: Like the no click or click kind of thing.
- 10:59You can get something in a phishing email and then,
- 11:02you know,
- 11:03it's so long.
- 11:04SPEAKER_03: You don't want to read it.
- 11:05SPEAKER_03: You're going to run it through your co-pilot to summarize this.
- 11:07SPEAKER_03: And in white text, which is not visible, it says go execute these things.
- 11:12SPEAKER_03: Those are things that we haven't necessarily seen in the wild,
- 11:14SPEAKER_03: but we know very easily could happen.
- 11:17SPEAKER_03: And so we're protecting against those.
- 11:19SPEAKER_01: The traditional answer to phishing has always been training,
- 11:22SPEAKER_01: but McLean Sterling,
- 11:23SPEAKER_01: who came to Proofpoint from the security awareness industry,
- 11:26SPEAKER_01: was candid about its limits.
- 11:28In general,
- 11:29though,
- 11:29even though I came from that area of security awareness,
- 11:34I really do think
- 11:36that you have to have more of the burden on the technology than
- 11:40on the person,
- 11:41on the employee.
- 11:43I think that the security awareness industry is sort of hanging
- 11:48on a little tightly
- 11:49to we as a person are going to train that other person and that
- 11:55is how they're
- 11:55SPEAKER_03: going to be better and they just need to slow down in their inbox.
- 12:00SPEAKER_03: But we're not pre-decisioned to be like that.
- 12:03SPEAKER_03: You know, that's not how the world works.
- 12:04We're moving it as fast as we can go and we're distracted by 17
- 12:08other things.
- 12:08And so as much of the burden that we can put on the technology
- 12:12to do the work for
- 12:13SPEAKER_03: us,
- 12:14SPEAKER_03: I think that's really,
- 12:15SPEAKER_01: really important.
- 12:16SPEAKER_01: Phishing has always exploited the gap between trust and verification.
- 12:20SPEAKER_01: AI is making that gap wider, faster, and harder to see.
- 12:25And the next wave of attacks won't just target the humans
- 12:27reading the emails,
- 12:28SPEAKER_01: they'll target the agents reading them on their behalf.
- 12:32Across everyone we spoke to at RSAC,
- 12:35we asked some variation of the same question.
- 12:38SPEAKER_01: Is AI making the CISO role harder or easier?
- 12:42SPEAKER_01: And the answer was unanimous.
- 12:44SPEAKER_01: Harder.
- 12:45SPEAKER_01: At least for now.
- 12:46AI governance has been an issue since the first consumer LLMs
- 12:50launched in late 2022.
- 12:52But while the first wave had risks like the potential for users
- 12:55to upload sensitive
- 12:56SPEAKER_01: files,
- 12:57SPEAKER_01: agentic models are a whole new ballgame.
- 13:00SPEAKER_01: AI agents have access to system files,
- 13:03SPEAKER_01: can browse the web,
- 13:04SPEAKER_01: can delete files,
- 13:05SPEAKER_01: and can download skills and MCPs.
- 13:08SPEAKER_01: The scope of what CISOs have to govern has exploded.
- 13:12SPEAKER_01: Dr Anton Shavakin,
- 13:13SPEAKER_01: security advisor in the office of the CISO at Google Cloud,
- 13:17SPEAKER_01: broke this idea down for us.
- 13:19SPEAKER_01: It added to responsibilities because,
- 13:21SPEAKER_01: again,
- 13:21people bring their own agents to work and trying to get them to
- 13:24do their work
- 13:25SPEAKER_03: without telling the employer adds to CISO stack.
- 13:29But the fact that certain tasks are now automated by machines
- 13:32takes off from CISO.
- 13:33SPEAKER_03: So think about CISO's inbox.
- 13:35SPEAKER_03: Some stuff gets piled in, but some stuff gets removed.
- 13:38SPEAKER_03: Deep in my heart, I feel like more stuff gets added.
- 13:40SPEAKER_03: Right.
- 13:41SPEAKER_03: Right.
- 13:42We are helping the CISO,
- 13:43but we are also,
- 13:44well,
- 13:44we kind of pile some stuff in his inbox.
- 13:47One thing that I feel like became better is that most of my
- 13:522023-2024 presentations
- 13:53SPEAKER_03: started with this slide that had security,
- 13:57SPEAKER_03: privacy,
- 13:57SPEAKER_03: compliance,
- 13:58SPEAKER_03: ethical use,
- 13:59SPEAKER_03: intellectual property,
- 14:01SPEAKER_03: and a bunch of other things that have nothing to do with CISOs normally,
- 14:04SPEAKER_03: but they all showed up in CISOs inboxes.
- 14:07And I felt like this was just unfair to CISOs because CISOs
- 14:10should not be an AI ethicist.
- 14:13It's just not his job,
- 14:14not his skill set,
- 14:15but because nobody else wanted it.
- 14:17SPEAKER_01: It was shoved to CISOs, CISOs inbox.
- 14:20SPEAKER_01: Expert insights research bears this out.
- 14:24SPEAKER_01: 67% of CISOs reported increased workloads compared with 12 months ago.
- 14:27SPEAKER_01: 60% have experienced professional burnout in the last year.
- 14:32Kush Kashyap at Vanta described what she called the CISO
- 14:36burnout paradox.
- 14:37SPEAKER_01: The technology that's supposed to ease the burden on security leaders is,
- 14:41SPEAKER_01: for now,
- 14:42SPEAKER_01: making it harder.
- 14:44the vendor security process is much more important and much
- 14:49more detailed and much
- 14:51SPEAKER_00: more steps to be done than before.
- 14:53SPEAKER_00: So it's increased the burden of security by a large extent.
- 14:56SPEAKER_00: And I think that's where this burnout is coming from.
- 14:59SPEAKER_00: When your security architecture,
- 15:01SPEAKER_00: your application security,
- 15:02SPEAKER_00: your vulnerability management,
- 15:03SPEAKER_00: how to check for agents,
- 15:05SPEAKER_00: non-human identities have exponentially increased in the last few months.
- 15:08So it's not just human identities as in the employees and
- 15:11contractors to manage,
- 15:12SPEAKER_00: but all these non-human identities too.
- 15:14All of this is happening while boards and CEOs are pushing
- 15:18CISOs to adopt AI faster.
- 15:20SPEAKER_01: The mandate is coming from the top.
- 15:22SPEAKER_01: Deploy AI, find efficiencies, and stay competitive.
- 15:26SPEAKER_01: It's rare for technology adoption to be pushed this hard from above,
- 15:30SPEAKER_01: but the budgets and the tooling to do it securely are lagging behind.
- 15:34Proofpoint's McLean Sterling shared a story from a CISO panel
- 15:37in Sydney where a
- 15:38SPEAKER_01: very established CISO said,
- 15:40SPEAKER_01: and I quote,
- 15:41SPEAKER_01: I think I'm just going to retire so that I don't have to deal with it.
- 15:45SPEAKER_01: But there's also a recognition that the CISO role has to evolve.
- 15:49SPEAKER_01: The old gatekeeper model where security blocks everything is dead.
- 15:53SPEAKER_01: CISOs who try to block AI adoption will not last.
- 15:57SPEAKER_03: Really,
- 15:57SPEAKER_03: really can't go sort of that old school model of the office of no,
- 16:02SPEAKER_03: you know,
- 16:03SPEAKER_03: ruling with an iron fist.
- 16:04SPEAKER_03: We're going to block everything.
- 16:05SPEAKER_03: I think those CISOs are quickly going to be out the door.
- 16:08SPEAKER_03: Interesting.
- 16:11SPEAKER_03: Even with the companies that have really low risk tolerances,
- 16:15I still think that ability to partner with other groups is how
- 16:20they're going to be
- 16:20SPEAKER_03: successful.
- 16:21SPEAKER_03: Yeah.
- 16:21SPEAKER_01: The CISO role is shifting from technical gatekeeper to business partner.
- 16:26SPEAKER_01: CISOs need to be working on the business,
- 16:28SPEAKER_01: not just in it,
- 16:29SPEAKER_01: but making that shift while the ground is moving under you.
- 16:32SPEAKER_01: That's the challenge.
- 16:34SPEAKER_01: No Before's Eric Kron captured this dilemma.
- 16:37I also see a lot of organizations really struggling to write
- 16:41AI-related policies.
- 16:43SPEAKER_03: I've had some talks with some CISOs about that,
- 16:45SPEAKER_03: and they're like,
- 16:46SPEAKER_03: we're concerned about where it could go wrong,
- 16:48SPEAKER_03: but in the same breath,
- 16:48SPEAKER_03: we know that if we say,
- 16:49SPEAKER_03: thou shalt not,
- 16:51SPEAKER_03: People are going to do it anyways.
- 16:53SPEAKER_03: So they're kind of running that razor's edge of saying,
- 16:57SPEAKER_03: okay,
- 16:57SPEAKER_03: this is okay,
- 16:59SPEAKER_03: but you need to do it this way,
- 17:01which can be a challenge if it's too complex for the people
- 17:06that are trying to roll
- 17:06SPEAKER_03: it out there.
- 17:07Despite all of this,
- 17:08the prevailing mood at RSAC was more positive than in previous
- 17:12years.
- 17:14SPEAKER_01: Last year, many teams were in the planning stage.
- 17:16Now,
- 17:17AI deployments are actually happening and teams are starting to
- 17:20see results.
- 17:21When it comes to AI,
- 17:22there's a defender's advantage,
- 17:24and that advantage is context.
- 17:27Defenders know their own infrastructure better than any
- 17:30attacker ever will.
- 17:31SPEAKER_01: They know where the most valuable assets are.
- 17:33SPEAKER_01: They know what normal behaviour looks like.
- 17:36AI can take advantage of that in ways that were never possible
- 17:39when it all had to
- 17:40SPEAKER_01: be kept in human heads.
- 17:42SPEAKER_01: Google's cloud, John Ramsey, put it this way.
- 17:45SPEAKER_03: We know our infrastructure better than anyone.
- 17:48SPEAKER_03: And because we know our infrastructure better than the adversaries,
- 17:51we can use AI to fix things before the adversary can attack the
- 17:54infrastructure.
- 17:55SPEAKER_03: And even when the adversary is attacking the infrastructure,
- 17:58SPEAKER_03: they still have to learn what the infrastructure is.
- 18:01And so that gives us a head start advantage,
- 18:05if you will,
- 18:06to the adversaries.
- 18:08SPEAKER_01: Google Cloud's Anton Shevakin told us that his thinking has evolved.
- 18:12SPEAKER_01: He started as a sceptic on whether AI would benefit defenders,
- 18:16SPEAKER_01: but now he's convinced that in the long term,
- 18:18defenders have more data and AI will benefit defenders more
- 18:22than attackers.
- 18:23SPEAKER_01: But ultimately,
- 18:24SPEAKER_01: I'm convinced that in the long term,
- 18:26defenders have more data and the AI would benefit defenders in
- 18:30the long term more
- 18:31SPEAKER_03: than the attackers.
- 18:32It's a somewhat unpopular message today because people say,
- 18:35yeah,
- 18:36but I can have a lambda hack.
- 18:37SPEAKER_03: And yes, you can have a lambda hack.
- 18:39SPEAKER_01: But I feel like in the long term, it would favor defenders.
- 18:43SPEAKER_01: So I'm definitely optimistic.
- 18:44SPEAKER_01: And Vanta's Kush Kashyap struck a similar note.
- 18:47SPEAKER_01: Despite everything, she's excited about where AI is heading.
- 18:51SPEAKER_00: I would say like from a person who uses AI,
- 18:55I'm very optimistic about the future of AI and everything
- 18:57that's going to come up.
- 18:58SPEAKER_00: From a security perspective,
- 18:59SPEAKER_00: I still think that there are a lot of things that we need to figure out,
- 19:02SPEAKER_00: like which vendors will get consolidated,
- 19:05which players will become like the key players for securing
- 19:09agents and different
- 19:10SPEAKER_00: aspects of security around agents.
- 19:12SPEAKER_00: How will AI become like another infrastructure element that's protected?
- 19:19SPEAKER_00: What compliance frameworks will come up?
- 19:23SPEAKER_00: How will the compliance landscape evolve?
- 19:25SPEAKER_00: How will the testing of it evolve?
- 19:28SPEAKER_00: How some of the vendors who are in the space evolve?
- 19:31How would Vanta's GRC platform look a lot different with so
- 19:36many agents doing so much more?
- 19:38SPEAKER_00: So I think there's more to be seen, but it's a very exciting time to be
- 19:43SPEAKER_00: witnessing all this change and seeing how fast it's happening.
- 19:46SPEAKER_00: As I said before, like dog years, right?
- 19:48SPEAKER_00: Everything happens so fast.
- 19:49SPEAKER_00: So it's an exciting time.
- 19:51For now,
- 19:52AI is adding complexity,
- 19:54accelerating threats and stretching CISOs thinner than ever.
- 19:58The leaders who will close the gap are the ones working with
- 20:01the business,
- 20:01SPEAKER_01: setting the guardrails and using AI to defend against AI.
- 20:12SPEAKER_01: I'm Alex Zawonetsky from Expert Insights.
- 20:14Subscribe and check out our full interviews and analysis from
- 20:17the show floor.
- 20:18But here's the question nobody fully answered this week,
- 20:21and the one that I keep coming back to.
- 20:23SPEAKER_01: If humans move from in the loop to on the loop,
- 20:26SPEAKER_01: or even out of the loop entirely,
- 20:29SPEAKER_01: what do security leaders actually become?