EP221 Special - Semi-Live from Google Cloud Next 2025: AI, Agents, Security ... Cloud?

Security warning. Hi there. Welcome to the Cloud Security Podcast by Google. Thanks for joining us today. Your hosts here are myself, Timothy Peacock, the Senior PM for Risk Analytics and Detection Things in Google SecOps, and Anton Chuvakin, a reformed analyst and senior staff in Google Cloud's Office of the CISO. You can find and subscribe to this podcast wherever you get your podcasts, as well as at our website, cloud.withgoogle.com/cloudsecurity/podcast. If you enjoy our content and want it delivered to you piping hot every Monday, please do hit that subscribe button. You can follow the show, argue with us and the rest of our Cloud Security Podcast listeners all on the Google Cloud Security Community page. Link to that is in the show notes. Anton, this is a rebroadcast of something we did live from Google Cloud Next 2025. And boy, was I sleepy in the first half of the episode.

Well, I wasn't even there for the first half of the episode and frankly, even for the second half. Ah, that's true.

You were remote.

I was remote and you were there. So I think it qualifies as live, but it's also kind of semi-live, I think maybe.

Semi-live. Yes, that's right. And, you know, it's so hard to do these when you're not there to bounce the ideas off of me. And it felt really lonely to be there. And I have to say, a lot of people said, Tim, where's Anton?

Okay, this made my day. And this makes me feel good about recording this opener. I think it's I had a chance to experience Next through your eyes, which was kind of good.

Yes.

I still, at the time of recording, I haven't watched any videos yet except for the keynote. I did watch it live. And you remember that there was this moment during the show when I was genuinely inspired. We should just air it for that moment alone.

Yes. Okay. Well, listeners, for Anton being genuinely inspired, let's turn things over to ourselves.

We sure do.

Good morning, listeners. We are coming to you live from the show floor at Google Cloud Next 2025. I am in the Mandiant Cafe here at the Google Security booth, and it's packed. Look at all these people. Look, there's a ton of people learning about security, Anton. This is the busiest I've ever seen our booth at any of these shows.

Oh, wow. That is impressive.

Yeah, I don't know what we did this year. Maybe it's like they put the screens very close together so it feels dense, but there are undeniably A lot of people here today.

So let's do our sort of like interview slash discussion slash recap. I don't know what we are doing. We are having fun in 2 different places here in Vegas.

Listeners, we don't know what we're doing today. This is a very chill episode.

Yeah, this is a very chill episode. So any themes? The usual question I get asked about these things is, what's the theme? What's the theme? So Tim, you're there. What's the theme? Is there a theme? I mean the security side, not just like AI.

It's AI.

Oh, it's AI. It's AI.

Everywhere the theme is AI. Yes, it is Gemini this, Gemini that. It is AI all over the place here.

And I assume that means securing AI, using AI for security, and AI in all other forms, shapes, and colors.

Yes. Yeah, yeah, yeah. So we've got the Model Armor announcement and that's cool, but I don't know a lot about Model Armor, but I don't have the impression that it is AI for AI security. I have the impression that it's security for AI.

Yes, it's definitely the approach to secure AI on GCP and a few other places. So, that part is there. So, let me ask you a strange question. Last year, when we were doing live show from Next 2024, I kind of had a bit of a similar vibe. There was a lot of AI. There was a lot of AI in the keynote. There was a lot of presence on AI. I did the presentation on AI. So, what is different with AI this year? Because that felt like 2024 was about AI as well.

Yeah, I think right now it feels a lot more real. Last year, AI felt a little perhaps premature or excitement-based. And I think this year everybody I've talked to has had some real experience with the thing and has really seen how it's going to change the world. And if you'd asked me a year ago, Tim, will there be SOC 1 analysts in 10 years? I would've said yes. If you ask me that same question today, I would say no way.

Okay. So, that is a change. And I think that's a change that you can touch. I remember it's a— we're not doing a show on RSA, that's in a month. But I remember last year's RSA, somebody was really annoyed with all the AI. And I said, we are in the use case discovery process.

Yes.

So, it sounds like we were in the use case discovery process at Next last year. And now, we've discovered some use cases. And now, we are proceeding down a few paths, right? Is that a fair assessment?

I think that's very fair. I think we are— We're somewhere in between having left discovery and having really hit the fat part of the adoption curve. And I think it'll be really interesting to see which of the trains that are steaming out of the station make it to the next station and which of the trains don't.

As a completely unrelated bit, yesterday I asked Gemini to give me a few AI startups that are likely to succeed. And 2.5 gave a very nuanced answer.

Oh, that's fun. What did it say?

Yeah, I didn't ask, give me a fun AI startup. I asked more like seed, and I got a pretty nuanced analysis from 2.5 about why it thought these 5 have a higher chance of success. It's really fascinating. It's almost like asking AI about success in AI security produces good results. And obviously, every vendor can do the same, so I'm not revealing the list.

Did it name real startups, or did it hallucinate them?

No, they were all real, for sure. And the answers were quite robust.

I mean, really, they were all real. Did you check?

Yes. I mean, I knew 80% of them, so 4 out of 5, and the 5th one I did not know, but I looked them up immediately. So yes, they definitely— oh no, we are beyond making companies upstage. Pretty clear. Yeah, that's to 2023.

Are we? Are you sure?

No, I'm not sure. I'll be honest with you.

I don't know if I trust you myself.

Well, I mean, I did look at Google.

I definitely trust Google a lot more, but But that's well known. See, you looked it up.

Yeah, I definitely looked it up. So I think the other question I prepared about the single most impactful thing is kind of maybe you're going to go to AI, but let me still try. In terms of security announcements, in terms of security news, what was the biggest thing that maybe impressed you the most? Like, I'll combine the questions. What's the biggest? What impressed you the most?

Oh, what's the biggest? I think the biggest has to be Gus. I mean, we had the GUS demo on the main stage at the keynote.

Yes.

And it's pretty special to have that kind of visibility and opportunity. And listeners, the other fun thing is this is an opportunity for me to plug this. I was the understudy for the demo during the keynote, and it was pretty cool to be backstage during all of that, see how that kind of production comes together. And I worked really hard to do absolutely nothing.

But you assured the success of the keynote by being there, right?

I did.

You did, yes.

And it's not quite true that I did nothing. During the keynote, while Payal was on stage, while my skip level was on stage, listeners, I was backstage working a second laptop, mirroring what she was doing, just in case her laptop on stage caught on fire, they could have cut over to mine live. And we would have had a backup there. And then there was another backup for that, which was a recorded video of the demo.

Yeah, the keynotes are really well prepared. They have that vibe.

So they had 2 nets under her on stage.

I watched the keynote. It went flawless. So it was very clear that this is a lot of hard work.

Yes.

And I mean, believe it or not, I did watch the whole thing from start to finish, not only the security parts. And I was kind of impressed with how much we've done. I mean, I know keynotes are meant to produce this effect, But we are pretty jaded security people. So, I felt very inspired by being at Google. Like, I'm not joking. I'm not doing cynical Anton remarks. I felt very— I'm happy that I'm at Google.

Listeners, listeners, listeners, Anton felt inspired. For Anton to feel that way, this is a big day.

It's a pretty freaking high bar. I would go with that. Yes.

It's a really high bar. It's a really high bar. Did you see the flowers demo? The flowers demo was what impressed me the most, though, of the keynote demos.

Oh, yes. Flowers demo blew my mind. Yes.

That was real, by the way. That was real. Yeah, they did that. actually talking. And during the practice runs, it worked to varying degrees. And so while that was happening on stage, all of the speakers were in the speaker room just holding our breath, biting our nails. And to see that succeed as well as it did on stage that day, we were all delighted.

Yes. And I think the pauses— I personally freaked out during some of the pauses when it was thinking about giving the discount. Yeah, it's actually kind of mind-blowing, like It's like the thinking was palpable. Maybe it's like API delays, whatever. It may not be thinking, but it was very clear that the system has magic.

It's pretty cool. So, it's kinds of experiences like that that make me say AI.

Yeah. We wanted to go kind of back to our podcast title, the cloud security.

Yes.

And last year at Next, we had this cafe conversation, I think. I think it was in the cafe about peninsula versus island. What it was is whether security, cloud security, is kind of built into the overall body of security, cybersecurity, information security, or it's still an island. Like, you are an endpoint security expert. Yeah, I'm a cloud security expert. We are not friends. So, this whole peninsula island thing, I felt is kind of impactful. But at the same time, this is not an underhanded way to get you to talk about Wiz. I'm genuinely curious, what are we changing? What is changing in terms of cloud being an island or a peninsula, part of the overall security story?

I've talked to a lot of users here this week. It is not yet part of the mainland.

Yes, that's very—

What I hear is that cloud remains an island.

Yes.

And I think it remains an island because of human nature and because human beings want to specialize in things. It's pretty natural. But I think what I hear from SOC teams and what I hear from defenders and Program teams, people specialize in cloud, and they're not cloud plus networking, they're not cloud plus EDR, they're not cloud plus identity on-prem, they're cloud. And I think cloud security is going to remain this way, possibly for a long time. I mean, Anton, are there other examples in security where domains became part of the mainland, or do domains tend to remain like the— Security archipelago.

I think we— that's also— I think I'm having a déjà vu from last year's Next when we were in Vegas, because I think mobile security is an example that is less separate these days. Like an Android endpoint is an endpoint.

But also less present.

Much less present. Correct. Correct. Yeah.

Yeah.

Okay. So this makes sense. I think I agree with your assessment about it being an island. And it's not about good and bad. It's about that the approaches to cloudsec are often magical, special, and different compared to, like, the rest.

And it's a deep domain.

Yeah.

It is a very deep domain to get cloud right. And it's not surprising that people can only hold so much stuff in their head.

There is that. So, let me ask, I mean, I promise it's not a leading question to highlight, you know, your involvement in Keynote and all that stuff. But we do talk about shifting security burden from humans, from clients, from, like, analysts to sort of platform features and automation. So, have you seen examples, whether they're inside Gus or outside, where we do that, like, where we continue to do that?

You know, I haven't seen a ton of crazy, like, governance burden shifting features, but we're making a lot of noise about confidential right now. And I think with AI, confidential is kind of having a moment. So, we're making a lot of announcements there. But to be honest, in the space of how do I get my posture right, how do I get my VMs patched, how do I redeploy and rebuild my containers faster, I haven't seen a lot that's going to take the burden off of good solid operational muscle.

Yes. I think that the— but it's also maybe the area where we see more evidence that the magic is coming, that it's more likely to happen. Rather than who knows, aspirational future, blah, blah, blah. Like, this is— the path is more clear to more changes, right? That to me is— I'll go with that.

Yeah. And I do think a lot of the things that we're doing manually today, we could totally do via agents in the future in cloud.

Yes. And I think that cloud specifically is where this may happen because all this like Snowflake, single servers built by hand from the '80s, I don't fully trust that agents will happen there anytime soon. Yeah, I'm sorry, startups that are building agents for mainframes. I don't think it's going to work out for you.

If I find that startup, I will let you know.

Right. Thank you. Well, agents to improve patching in Windows 2008. Oh, wait, that's out of patching window. So, no agents for that. Yeah, okay, fine. I mean, automation implies modern systems.

I'm slightly hungover, listeners, but that made me feel very queasy.

Yes, I agree with that. And I think that the reason why some of our stuff internally is quite magical is because our systems are more modern than many others. In that sense, it's easy to do some stuff. Zero trust, right? I'm pointing at the old BeyondCorp logo on my shirt here. It worked out for us because the systems are aligned with it. Okay. I'm departing too far from the theme. So sorry for—

So no, but it's good you brought up beyond trust. I actually had a really interesting conversation with a security leader yesterday talking about the company's efforts on zero trust. And I said, well, are you starting with machine identities or human identities?

Yeah.

And you know what he said?

Nope.

He said both.

Starting with both is usually wrong because if you start with both, you end up committing to too much, right?

That's what I tried to tell him. But this is a very, very capable security leader I've known for a long time. And so if anybody's going to pull it off, he'll be the one to do it.

OK. That's fair. I think that makes sense to me. So you are in the vendor expo. So what about the expo floor buzz? What are people trying to sell you?

Oh, they're trying to sell me AI, Anton. Everything here is about AI. This episode could have been really short. We could have just gotten on and said it's all about AI and hung up. But there's a lot of things here for AI. Equinix is offering AI-ready data centers. Vendors are offering AI security. We're offering AI security. It's AI everywhere.

Yes, I think that's kind of— I mean, that's an easy answer, but it also with a predictable and maybe not boring, maybe it's an exciting question, but it's a predictable one. I dread RSA expo floor in a month when we would see What is it, 800 vendors all selling AI to you? That would hurt.

So I, I want to talk about Wiz for a second because their booth didn't say AI.

Wow.

So Wiz's booth, I think, is a really interesting thing. Imagine you're a marketing person. What do you do to market a company that everybody at the show has already heard of and already knows everything about? Do you put CSPM, GWPP, CNAPP? What do you put on the side of the booth?

I mean, if everybody knows you already—

They didn't put any of that. You know what they put on their booth?

Yeah, what?

They put, and I kid you not, they must have 100 different fake DVDs of popular movies which were twisted to be security references.

Whoa, take a picture please. I want to see that.

Yeah.

I think last year's, their booth was quite amazing with all this.

I will send you the picture that I've taken.

Yes.

So it's as if they took the food thing and redid it slightly differently with a bunch of movies where every movie has been warped to be a security reference. It's brilliant.

It is brilliant. Oh, wow.

Yeah. Yeah. So Finding Nemo, Patching Nemo.

Patching.

And then like the jokes aren't that funny, but there's a lot of them.

A lot of them.

Let me see. I took a photo yesterday. Let me see what I got. But that's the funny thing is the jokes, they're really not that funny. Like they're just security jokes. Forgetting Sarah Marshall's keys. Not Mean Girls, Mean CISOs.

Mean CISOs, yes.

V for vulnerability.

V for vulnerability.

Not the Karate Kid. The Terraform kit.

The last one is cringy. I'm sorry.

It's cringy. They're cringy. They're cringy jokes. Like they're not, they're not good jokes. You don't look at these and go, ah, you look at these, you're like, you guys made 100 of these weird security jokes instead of talking about what your product does, because everybody here already knows what your product does.

Everybody knows. Yes. Yes.

And I think that's a really interesting marketing move. And I can't imagine most organizations having that courage. It's a gutsy move to do what they did.

I would agree with that because ultimately, this marketing magic is not necessarily in spelling out CNAPP better than others. Some of the magic is in the attention. Marketing is psyops, right? We all know that. Like, good— well, I guess good marketing is psyops, right? And so, that qualifies. That is pretty amazing. Okay. So, let's go to surprises a little bit. So, I guess I'm trying to— I think that qualifies what was the biggest surprise at the show for you, and maybe a vendor booth without AI on it was a surprise. But anything more tangible? Any surprises that struck you as, what? Is this really happening?

Well, I do want to talk about how crowded our booth is. I mean, it is packed over here in security land, and people are are genuinely curious about what's going on. There's a lot of buzz. And so, I am honestly surprised by how interested people are in that stuff.

I mean, Gus announcement is an interesting bit because ultimately, it unifies some things that maybe others didn't dare unify. So, we'll see how it fares. But I was kind of impressed with some of the thinking behind Gus. Do you sense any— I don't want to turn this into announcement show, but This was an interesting one. So any reactions in the audience? In the—

You know, I was in the basement when the Gus demo happened, so I don't know what the audience said. What have you heard? What— or maybe let me ask you a question. What about Gus is surprising to you?

Oh, you want the non-cynical answer, right? Or you want the cynical answer?

Well, I got inspired, Anton, a minute ago, so let me continue to get the non-cynical answers.

Yeah. OK. So I think that the logic behind— combining SecOps magic that we clearly have, and some of the very unrelated magic, like the fact that Chromebooks and Chrome OS is very secure and is not vulnerable to anything that plagues traditional systems, is a good bridge. It's a bridge between 2 things that are amazing, but the bridge itself is a little difficult to build. And I think we managed to mostly build it. That, to me, is the positive part, is that some of the DNR magic, some of the Intel magic, and some of the secure system design magic got bridged in the form of GAS. Whether this bridge would hold in every environment, including environments with Windows 2008 still running, I don't know. But it will run— the bridge would hold in modern environments and those aspiring for modernity. Maybe I'll put it this way.

That's fair.

I think that's how my ex-analyst brain is analyzing the GAS announcement is that it's It does have the magic. It does have multiple sources of strength that we have. And they're connected in a way that would hold in some places, but perhaps not in others.

I think it will be interesting to see how we adapt and evolve Gus as it goes forwards. But I will say, at least internally, a lot of people are very excited for us to be riding the Gus bus.

Riding the Gus bus, yeah. The Gus bus.

Yes.

Okay, so did you make this joke in advance or is it an impromptu one?

Everybody has been making the Gus Bus jokes.

Oh, it's just I'm just missing out. I'm not connected to the cool kids. I get it, I get it.

Well, now you are.

Now I am. Now I can make the Gus Bus jokes. Yes.

Yes, now you are.

Because my mind was kind of going to the flight of a goose or something, but it was definitely very off.

No, it's not a goose.

It's a Gus.

It's a Gus.

Okay, so—

I will say, this is a funny thing. I'll give you a backstage secret. We had a lot of trouble with how TK was pronouncing the different names of the products, and people had to tell him, "Hey, it's actually pronounced this way." But our products are used; they're all descriptive names.

Hmm. Okay.

Did you watch the keynote? There's yeah, imagine there's VO.

Oh yeah, yeah, yeah. Not AI, not security products. Yeah, yeah, yeah. No, that part is I know that I'm with him on this one because I. I imagine, imagine, I would be like, what? Okay, I know what the product is, but I don't know exactly whether it's Imagine or Imagine or Imagine.

Apparently it's Imagine. I also got that one wrong.

Oh, well, there you have it. So we are with TK.

There you have it.

So let's go to a few other fun things we have. So let's predict. Let's predict.

Yes, please.

This is 2025.

Let's predict.

Imagine.

Our favorite game.

Imagine. We are at NEXT 2026 in Vegas, or maybe one of us is in Vegas, who knows? And we are discussing how things went. So what do you foresee today? What do you imagine to be there in 12 months?

What do I imagine to be here in 12 months? Well, more AI for sure.

Even more AI. By this you mean even more AI?

I think we will start to have talks even more AI. I think by next year we will have multiple customers and multiple vendors doing sessions talking about how they have changed their SOC throughput and entire SOC management philosophy. Yes, with agentic approaches to triage.

Okay, you think 1 year is enough for that? Honest question.

Oh yeah, oh yeah. I'm kind of surprised we didn't have it already, to be honest.

I'm not, because a lot of messy stuff, so a lot of messy stuff in many socks. Some of it is kind of hard for machines. A lot of, oh yeah, I know John knows, and then John knows that actually Joanna knows on the second floor and you have to call them. Agents cannot do that yet. So, a lot of tribal knowledge is kind of hard to handle. I guess in theory they can.

I think the agent will learn that pretty quickly.

Yeah, you're right. Fair point. It's not impossible for them to do that. They just don't do it yet. I, for one, would be super curious to see that future. I mean, there are many areas of security where In a modern environment, agentic stuff would have a lot of impact. And so, one year is enough for that impact to be seen, internalized, and then presented as a success story, not as a promise. I think that's maybe— I'm going to watch some of the speeches. I've started doing this. But I noticed that some of the speeches clearly show promise as opposed to celebrate the results. And I'm all for promise, but I also want results.

Yes.

And in a year, I think results would overpower promise.

Results are better than promised.

Results are better than promised. And we will have much more results in a year than this year. That's fair.

Yes. So, that's my prediction. My prediction is a year from now, we have people talking about genuine SOC transform. Yeah.

Based on agents, specifically, not just SOC transformation, but specifically based on agents, agentic-powered SOC transformation. Yeah. I'm super curious to see that. Yeah. And again, I believe That's right. A year may be enough, but I'm hedging a little bit with May.

What do you think you're going to see in a year? I want your prediction too. I can't be the only one out here predicting.

Yeah, of course. Well, again, I'm not at the show. I'm only looking at some of the speeches and, of course, the keynote. I think there will be sizable areas where AI have changed things in real life. You're right. connect to your prediction and say that, as of today, if I look at startups, if I look at large companies, if I look at success stories, a lot of them are promise-powered. The promise is very load-bearing. They say, we've seen this, and the promise is this.

Yes.

Good. But people who tell me that when they applied AI of this type in this area of their security, operations, systems, engineering, this big result happened. Kind of like what we published, some of the papers about Google internally doing things, and the results are very clear, very measurable. They may be not 10x'd, maybe they are plus 50%, but I expect a lot more of that. I think that this year is going to be a shift from promise to results year. I think that the previous year was a shift from the types of promises, more promise, even more promise. I've seen some results, but I've not seen the results everywhere in the ether. I've seen promise mixed with a little results.

What comes after results? We've got idea discovery, we've got promise, we've got results, then what? What's the next stage in the analyst framework here?

Excellent point. I think that for after results, results is a range from slight improvement to bigger improvement. I think transformation is after results because if you have results, Oh. You may need to do things differently, or you may see the opportunity to do things differently. So, to me, transformation happens after results. You cannot have a promise-powered transformation. That would not happen. The friction would stop it.

Yes.

If you recall, somebody, another ex-analyst on the podcast, asked me something about my lessons from Gartner a couple of years ago. And I said that my one lesson from 8 years at Gartner was that IT inertia is the most powerful force in the universe. Yeah. And that I can almost always willing to bet on the past. I'm going to lose enough, but I'm going to win a lot more. So, for me to change that in a year, we'll see more results to do power transformation, it's kind of a strong position because betting on the past is a winning strategy if enterprise IT is concerned. Betting on the mainframe staying, betting on Windows 2008 being there.

That's one of the most depressing takes we've had on the show in a long time, Anton.

No, because change is hard and every human knows this deep down.

Yeah, that's true.

And a large organization change is harder.

I love the cartoon of who wants change? Everybody raises their hand.

Yes.

Who wants to change? Nobody raises their hand.

Or who is willing to work for years to change something complex with a lot of organizational friction? Human change is hard, but the organization, large organization change is much harder.

Culture change is very hard.

I think that's why I am inertia. I guess I would bet on inertia. And as I said, I'm going to lose enough. But overall, in enterprise IT, that's a winning strategy. All right.

Well, listeners, we've got the inertia lobby represented today.

This is great. It's a little— that's too depressing. So, please, you know. Okay. We have maybe one final question. What one thing should a security professional remember from the show? Yeah. Apart from, you know, AI matters. Or will matter, matter more.

Oh, yeah, no, I mean, it's really like if I had a single sentence, it is you must start planning for how AI will change every aspect of your business or you will be run over in 3 years' time.

Right. Excellent point about 3 years because you would be affected in a year, but in 3 years you'd be run over. Yes, that's exactly right.

You will be completely flattened in 3 years if you're not starting to think Really hard, really seriously right now. And that's true about everything. Like my fellow product managers, if you are not adopting AI tooling in your day-to-day as a PM, you're going to get flattened. SOC managers, you're going to get flattened. SOC operators, you're going to get flattened.

Yes.

So you might be betting on inertia, but I am betting on the steamroller of AI to flatten everything if people don't get out of its way.

I think that there is a way to actually combine these visions. They may be true because even in inertia-powered organizations, some things will get flattened by AI. It's very clear. So, I think that they're slightly conflicting, but ultimately they can both be true.

Yes.

Okay. Well, that's it. We're done. This was amazing. I miss Vegas. I guess I'll maybe be there later in the year. Come back, Tim.

Well, I'll see you at RSA. Listeners, you'll see both of us at RSA. Tune in for our RSA livestreams that we've got coming up. And Anton, I missed you in Vegas. It's not the same without you.

Thank you.

All right. With that, listeners, thanks so much for joining us today.

And now we are at time. Thank you very much for listening and, of course, for subscribing. You can find this podcast at Google Podcasts, Apple Podcasts, Spotify, or wherever else you get your podcasts. Also, you can find us at our website cloud.withgoogle.com/cloudsecurity/podcast. Please subscribe so that you don't miss episodes. You can follow us on Twitter, twitter.com/cloudsecpodcast. Your hosts are also on Twitter at Anton_Chuvaikin and TimPico. Tweet at us, email us, argue with us, and if you like or hate what we hear, we can invite you to the next episode. See you on the next Cloud Security Podcast episode.

Security.

Security.