EP275 Google Cloud Next 2026: The AI Earthquake, "SOC-home" Syndrome, and the Ragged Edge of Reality

Security. Hi there. Welcome to the Cloud Security Podcast by Google. Thanks for joining us today. Your hosts here as ever are myself, Tim Peacock, Group PM on Google SecOps, and of course, Anton Chuvakin, Senior Staff in Google Cloud's Office of the CISO. We are coming to you from Cloud Next 2026. This is our annual wrap-up episode. And this is our 4th cloud wrap-up episode. Is that right?

I think it's the 4th unless we missed a Next or 2 in the past. But yes, we did check. And this is the 4th time we are doing a recap. And we're going to reflect on some of the past recaps as well.

Do you want to start there?

Actually, let's start there. And I would also start from the whole security island metaphor. It was a really cool idea you had, I think, a couple of Nexts ago. And it was about how cloud security was born as a separate island, separate from the rest of security. So people produced cloud vulnerability scanning, cloud posture assessment, Cloud application security. Nobody made a cloud firewall, thank God. But I feel like the separation from the rest, and there was a cloud security team as well. And we felt like it was unusual because many other domains of security weren't born as an island. But cloud security was. And we felt like then by next year it reconnected and became kind of a peninsula. And maybe now the separation is less dramatic.

Really merging into the mainland.

Correct.

You really see like an Indian subcontinent, but without the Himalayas.

So maybe I need a different subcontinent. Yeah, it's a deeper metaphor, probably.

But what we're seeing this year is AI not as an island. We're seeing AI as a global phenomenon where every landmass is shaken by the emergence of AI, the release of agents, the capabilities of AI. So what does that tell us about the difference between cloud and AI?

It tells us a lot about the fact that Tim's metaphors are crisper than mine in most cases. And it just helps. I mean, your puns are better. So like, maybe this is my humble moment.

Wow, this is a special day for me.

Yeah, and your metaphors are crisper today.

Wow, today.

Today.

It's a good day. Got a good night's sleep.

But the AI is affecting the rest of security pretty much at the same time. So it is like a little bit of like an earthquake everywhere. I mean, at RSA, we saw people highlight the AI-native firewalls, whatever that thing is.

Okay, that made me laugh. But we have seen like AI-native email security.

We have absolutely seen that.

And I think we found that very credible.

And that seems really credible.

We've seen AI-native triage. We've seen AI-native SOC. We've seen AI-native malware analysis.

MDRs. They were AI-native MDRs that wouldn't exist without the current levels. That's right. And we've had them.

on the show too. And I think we found them credible. Correct.

And so I think this is actually an interesting point that it's not an island. And so maybe companies aren't hiring AI security teams, but AI security responsibility are diffused into other teams. There is a, say, AppSec for AI-powered applications and there are many other things. This to me is a fun one and it's kind of closing the loop on our next 2023, I think.

Yeah, well, 2023 was like we woke up to AI. 2024 was the island metaphor. '25 was the rise of AI. And I think '26 is the reality of AI. And so in 3 years, we've gone from, oh shit, there's AI, to, oh wow, there's AI everywhere and it's real.

I agree with 95% of that, but I feel like there is the whole ragged edge adoption. That is another metaphor.

Let's talk about that. Let's talk about the ragged edge of adoption. What does that mean?

So even here, even within the 2 days of Next, I've met people whose companies are just discovering AI and there's maybe a mandate from a CEO to use AI everywhere, but absolutely no governance. And I've pointed out that some use cases for AI, even in the US, would be completely illegal. And you're absolutely not supposed to do that. And people said, yeah, we know, but we're trying to convince everybody who's excited and who's heard the CEO say, use AI. And that was a smaller company. And again, I'm obscuring enough details so it's not any kind of—

But we also heard this week from gigantic 150-year-old companies that they're using AI at scale in production.

Correct. Correct. And one bit that happened during one of the previous recordings we've done is that Tim was surprised that not having a champion for AI adoption in security may hinder adoption. And completely, completely surprisingly, I felt like it's obvious.

Of course, that's obvious.

Earlier today, because during my other session with Allie, that was her first point about AI readiness that blew my mind. I want to talk about the data and the quality and the modernity of the SIEM APIs. And she basically said, Well, if the SOC, because AI adoption for a SOC does not have a person who is senior enough, understands AI, and wants to use it, your project will fail. And I was like, whoa.

What I heard that blew my mind even more than you need a champion is that there are SOCs that are resistant to AI, where there are people who don't want AI in their SOC and they're actively fighting to keep doing manual triage.

Yes, that was my goal.

I think they have some kind of SOC-Home Syndrome.

You're gonna win the pun battle today.

I am very proud of that one. That's a good one.

Yes.

And it's true. So aside from these people with SOC-Home Syndrome, I've heard basically every security leader I've talked to from planetary-scale food companies to government agencies to fintechs, they want agents. Everybody wants it. And it feels like there's kind of 2 things we're seeing in the SOC with agents.

Like why they want them.

Why they want them. There's kind of 2 categories of adoption. The first is to do things that we can do with humans better, faster, at greater scale. And the second is to bring completely net new capability into a SOC.

And by the way, net new does not mean, well, it's net new for them. Yes, because sometimes it's net new for everybody. Like Google may do things with AI that nobody has ever done and we're doing them and it's kind of amazing. But for some other companies, the take is really bizarre because they knew they needed to do it. They agreed it's a must. But they still didn't do it. Right.

So one example—

And now it became possible.

One example of category 2, this is the net new, is malware analysis. And so in Google SecOps, you can upload a binary privately to us. Our agent Gemini will decompile the binary, tell you not just this is good, this is bad, but it can offer you some really interesting insights. One user described previously $30,000 per malware analysis that they contracted out.

That's a human with a— we get kind of human—

A human contractor for a week.

Taking apart malware.

Exactly. And that reminds me of a funny story. from my childhood taking apart a camera and not understanding the difference between a battery. Anyway, it doesn't matter. $30,000 and a week of time to find out, no, this is run-of-the-mill malware. Now you can get that verdict in seconds from Gemini. And Gemini can give you insights like these are its C2 channels, these are the IOCs to add to your blacklists. There's a lot of capability all of a sudden that's brought to everybody through agentic capability.

And it costs essentially, I shouldn't make jokes and say it costs essentially nothing, but it costs essentially nothing compared to paying a human with a screwdriver.

And you don't have to wait a week. You wait 30 seconds. Yes. Yes.

That's a good point.

Yes. So incredible Type 1.

Any more announcements, by the way? This— sorry to be the talking marketing guy, but marketing people get paid for all that. What do they pay for exactly?

They pay for all kinds of stuff around here. Like, well, not our salaries, but our editing budget. So also GA of the Triage and Investigation Agent, which is a Category 1 capability to do things that humans do, but way faster. GA of Gemini Enterprise, including Agent identity, the training stuff, all of that. Model Armor to GA. We've got some really exciting stuff coming soon, like the Threat Hunt agent, which my good friend Nolan and John Murray demoed not too long ago. And if you watch the recording, it's got CISO Tim. I step into the CISO role for about 30 seconds. So there's a ton of great announcements. Are there any others that caught your eye?

I kind of felt mostly the SecOps and like broadening from agents in a SOC to agentic SOC to then agentic defense.

Yes.

I think there was a lot of fun stuff from Wiz, as I recall.

Yes. What did we see there? We saw colorful agents.

Colorful agents. Yes, correct. And I think that this would stick to the theme of red, blue, and somehow green.

Yes.

But I feel like this is one of those odd moments when it's very clear why Google and Wiz belong together. And I'm not being paid to say this by marketing, but of course I was initially skeptical. You're getting a little bit of marketing money. A little bit. But I was initially maybe not skeptical, but I was kind of not certain of all the reasons. But I feel like these examples are the beautiful, crisp reasons why So this is actually really good for everybody, including the customers, because—

Especially the customers. Now, I had a very depressing moment at the SecOps CAB. We were talking about Mythos and Vuln Discovery, and we had a room full of people. And I said, hey, who here, as a result of Mythos and Vuln Discovery powered by agents, is going to patch faster? Surely you're all going to patch faster, right? Can you guess how many hands went up?

Probably zero. Yeah.

Goose egg.

Because the reason for that, I can tell you right away from my Gartner days, is that most people have maximized it to the what they think of as an absolute limit. And it varies by system. It may vary by vuln. And that's why admittedly I was so miffed by our friends from, well, Anthropic when they published the blog after Mythos that required certain broad classes of vulns to be patched immediately. And this patching immediately thing is just not going to happen. Like no matter who says this, you're not going to patch enterprise database immediately and you're not going to patch a few other things immediately, no matter who and how loudly and backed by what facts would say it. It's just not done. So the logic I used, or the metaphor I used— and this one's crisp, I promise— was the kind of a sound barrier. If you have a propeller plane, you increase the power to the propellers, you floor it— well, there's no pedal, so you're like—

You push it all the way to the wall.

Yeah, you push it all the way, and you're not going to break the sound barrier because it just can't be done.

Right. Laws of physics.

Yeah, exactly. And this is kind of the example with passion, is that they can push hard. They can motivate people. They can beat people. They can pay people more. It won't happen.

So my extension to this, listeners, is as Anton and I were talking about this speed of sound of patching, is we actually did overcome the sound barrier. Chuck Yeager got into a jet plane, not a prop plane, put it into a dive, and went faster than Mach 1. On the other hand, the only captains to go faster than the speed of light are Captains Kirk and Picard.

Right, in sci-fi, of course.

In sci-fi.

I want to make it for people who watch sci-fi.

In sci-fi.

Too much Star Wars.

Yes. I did it. I did it. I— it's been 5 and a half years, listeners. I might have to break up with Anton after this. But my point is, are we looking at something which can be overcome by engineering and courage, or are we looking at something that is truly an impossibility? And I think for some customers, some situations, you know, imagine you're a petroleum refiner. You've got a SCADA system driving a fractional distillation tower. That tower works 24/7 for 3 years before you turn it off.

Yep.

You're not gonna patch during those 3 years.

It'll explode if you get it wrong. And moreover, there is some of the engineering-led solutions to re-architect, refactor are also not an option. So, I suspect in this metaphor, for some systems, it is kind of a speed of light. But for other systems, it is probably speed of sound because if you have an enterprise database, and again, I'll stick to that example, which takes traditionally about a quarter of patch, of testing, refinement, user testing, whatever else. So, like, it's probably on the slower side. Historically, it was very fragile, a lot of code written probably in the '80s. And you cannot speed it up.

Right.

But you can toss the database out of the window and write a new one, buy a new one, procure a new one, or store data somewhere else.

Right.

The point is that the re-architecting, re-engineering answer is there. And the new system would be either written securely, would not have the bounds, in question or can be rapidly patchable, can be architected in a modern way that would allow auto-patching despite the fact that it's a database. So the choices are there.

There are new choices.

But I fear that for Refinement Tower, there are none.

Right. And so I think the interesting thing is, you know, we've had decades of legacy tech where something was built by a vendor. That vendor is no longer contracted, no longer exists. So you're stuck with this unpatchable system. But it might be something basic that's just doing data processing or data entry. We're now in a world where almost any organization could use a tool like Antigravity to vibe code it and replace it with a modern stack.

I'd be putting some asterisks and caveats to this about almost any information services.

No, anybody can.

No, no, no. I think that you're right. But to use an example that I think we didn't use for the RSA podcast, if you bring me a GRC vendor that does compliance reporting, I think that your answer is yes, you can vibe code That. But if I bring you an EDR that works on Windows 2003, all the subsequent versions and has really tight, really well-written kernel code, which is really small and well-tested.

And that only causes a global outage once every 10 years.

And it only causes a global outage once every several years. You probably are not going to wipe code. I'm sorry.

Probably not.

I met people on Twitter who disagree with me and who think that, yes, they can do it.

Really?

Yeah, they can tight loop the testing and do all this, harness properly and basically do make sure that that happens. But admittedly, some of the Windows 2003 bugs may trigger only once in several years. So their agentic fleets of agents doing this would still miss it. So we're not going to rewrite a well-known EDR vendor by vibecoding.

No.

But one thing that connected to Mythos that came up is an interesting one. In completely unrelated circumstances, a couple of people who are not directly motivated for studying SecOps and SIEM in general basically said, Well, the vulnerability apocalypse, as we sometimes colloquially call this whole mythos acceleration—

Vulnpocalypse.

Vulnpocalypse, thank you. That works too. Would produce a kind of a newly recreated interest in doing detection response better, faster, would sell more SIEM licenses, would sell more EDR licenses, would sell more CDR, ADR, whatever other DR licenses.

When Anton first told me about this, my metaphor, and this is back to my crisp metaphors today, was we've got a wooden ship, we've got a new wood-eating termite. And somebody's solution is to sell us more buckets to bail out the water instead of maybe you should build your ships out of iron or steel or fiberglass or any other material.

No, a microscope to find termites.

Oh, no, that is the worst answer yet. So I think, listeners, what I'm trying to get at is, yes, we could use our old ways here and we will use our old ways as part of the bridge to getting there out of what somebody else this week called the Valley of Tears. But we also need to look at that rearchitecture. And unless we're doing that, I think as a security industry, we're not serving our business partners. Yes.

So the result of this and us thinking about it led me to think that, yes, there would be a boost in taking DNR more seriously, people deploying better visibility, doing more broad monitoring of whatever happens once your perimeter is breached by somebody with a zero-day. But you won't just DNR your way to success with failing prevention. You will have to rebuild systems, re-architect, replace, toss things out of the window and put secure alternatives in.

Yes. And then you still double down on detection and response where you need like your distillation towers and your other things that are very hard to patch or expensive or impossible. So speaking of expensive, we had a really interesting conversation with a CISO who thought of himself as also a CFO. And I had a whole panel talking about agentic metrics and how we're driving and moving the goalposts there. What do we want to say about that? Because I think that's a very interesting space.

So this was very fun. And I feel like the conversation with the CISO who works for kind of like a wholesale company, but used to—

Wholesale grocer with famously thin margins.

Right. And used to work for a bank and initially—

Famously thick margins.

Yes. Where the impression, initial impression was banks, money, they understand return, they understand value. But the reality is The lesson I learned from that conversation was that if you are really, really, really tight with budgets, you have like magic superpower to see value. He would not be sold crazy post-quantum AI mind-reading AIDR by a 3-day-old startup.

Right.

He would not.

He has a nose for it now.

He really has a nose for it because that nose would get slapped pretty hard if he makes a mistake. Sorry, now this is my metaphor. This is my metaphor win.

We're learning about Anton's dog training techniques.

I suppose that too.

But it is interesting how when you have effectively unlimited budget, you can end up with a lot of shelfware. You can end up with a failure to show ROI. But when you have a very thin margin, you as a leader have to think about what's the return here? What are my metrics? How do I prove to my leadership I didn't screw up with that purchase?

Actual business risks, not like a hand-wavy consulting paper about like security is a business risk, but it's a very like a juicy real-life embodiment of for him, security is very much a business risk.

You know what else he talks about in there? And he didn't use this word, But it was clearly he articulated a strong understanding of what his minimum viable business was.

Yes.

When we were talking about resilience, he knew exactly which processes he needed to first make sure worked, then second, then third, make sure it worked. And I think that reflects a degree of like operational maturity, mission focus, because, you know, delivering groceries, people will not eat if you don't roll the trucks. The ice cream will melt, the tomatoes will rot. Very interesting.

And he has another really good resilience lesson, which we can demonstrate right now. He demonstrated a digital device and said, okay, Okay, what if AI fails? And he immediately brought up a paper notepad and said, here's my backup. And it was very like, it's almost instantaneous. And he does think about like resilience, rollbacks, minimal viable business. That part was super impressive. I am sure banks do it as well. I'm perfectly convinced of that.

I'm sure, especially the ones that Phil worked for.

That's correct, where Phil worked.

But watch for this episode. It's coming out this week, next week. It'll be a midweek special because it's audio only. So if you've been joining us on video, we'll have it on YouTube. But if you've been joining us on audio for all these years, check out our YouTube. But this one will be another audio episode later this week. Do we have anything else?

This is very fun. And I feel like we want to revisit the ragged edge of adoption.

Ragged edge. Okay.

The reason why I want to revisit this as kind of a conclusion to this episode is that if you see certain reality around you and you think if you're on Twitter, you think everybody's using AI and they expect AGI by the end of the week. But if you are talking to very mundane old European businesses, you may think cloud is this new thing that will eventually come to them. So it is very— the ragged edge is very real and very dramatic. And the raggedness is kind of deep. Maybe something that you think was old in 2007, something that was seen as old in 2006, for somebody it's new in 2026.

Yes.

Which is kind of slightly scary, but it's also a reminder of how complex our world is.

So the future is here. It's just not evenly distributed yet.

Very.

This podcast is distributed. Wherever you get your podcasts, as well as at our website. And I guess, Anton, do you have a closing tip? Do you have recommended reading?

A closing tip is if somebody convinced you that everybody is doing X and you're not doing X yet, so you're a laggard, they're probably wrong.

I think Anton is preparing to be a father of a teenage girl and make sure that she doesn't do the things that teenagers do that they shouldn't do yet.

That's a useful reminder for the future.

That's good.

Exactly.

So listeners, with that, thank you so much for joining us today. Anton, thank you for a ton of fun this week at Next, and we will see you next week on the Cloud Security Podcast by Google. And now we are at time. Thank you very much for listening and of course for subscribing. Please subscribe so you can get new episodes piping hot. And if you love our content, please drop us a review on your platform of choice.

You can find this podcast on YouTube, Apple Podcasts, Spotify, or wherever you get your podcasts. Also, you can find us on our website, cloud.withgoogle.com/cloudsecurity/podcast. You can argue with us on the Google Cloud Security Community site, googlecloudcommunity.com.

You can follow us on X, x.com/cloudsecpodcast. Tweet at us, email us, argue with us. And if we like or hate what we hear, we can invite you to the next episode.

See you on the next episode of the Cloud Security Podcast by Google.

Are you ready? I'm ready.