EP169 Google Cloud Next 2024 Recap: Is Cloud an Island, So Much AI, Bots in SecOps

Security running.

Hi there.

Welcome to the Cloud Security Podcast by Google. Thanks for joining us today. Your hosts here, as ever, are myself, Timothy Peacock, the Senior PM for Risk Analytics and Cloud Threat Detection 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 on our website, cloud.google.com/podcasts. 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 on our community page. Anton, this is a fun episode because it's, well, it's just us. What do we say in an intro when it's just us?

This one is, well, we can have banter and then they're going to have more banter.

We can talk about how brilliant the guests are, perhaps.

We can do that. And oh, no, no, no. You know what? You know what?

What do we do?

Remember how this episode ended. How did it end? We tricked each other.

We did.

With the closing questions.

Oh, that's true, listeners. You can watch us be completely unprepared for something we've inflicted on over 170 people so far. You would think we'd be ready, and yet we were not. And so maybe without any further ado, wait for our foolishness till the end. Let's turn it over to ourselves.

That sounds kind of funny, but yes, let's do that. Turn it over to us.

Listeners, welcome back for our final livestream podcast of Google Cloud Next 2024. Anton and I have survived the desert. We've survived the joys of Vegas, and we have an exciting wrap-up to go through right now.

Now we're— We survived wearing jackets for a day.

By the way, look at us. You even wore a jacket for a day.

Look at us. We wear jackets. Googlers don't do that.

No, these are toxic to Googlers. We're actually going to have to go get therapy for our exposure later. But what I want to start out with is something I'm very proud of, which is something we invented together here at this table, the word freacher. What is a freacher?

Freacher is a feature that introduces so much friction by nature of its existence, it becomes almost not worth having the feature. So it's a friction-heavy feature.

Could you use it in a sentence?

Let me think about it. So, oh yeah, this button here to control whatever in this product, it's more of a feature because even though it produces the effect, it's so hard to use and so many other things need to happen that it's essentially a feature.

A feature. So we're very proud of this coinage. I think it's a good word. I encourage everybody to use it. Feature.

Feature.

Speaking of things that are not features.

No.

We talked about a lot of launches.

Our marketing people don't listen to the podcast. We know that because we've made jokes about marketing and nobody ever called us on this.

That's true.

Yeah. So we are pretty safe here, I think.

Yeah, we're safe here. But there are some good launches. We launched SCC Enterprise.

SCC Enterprise is kind of amazing because it officially made a lot of our security a lot more multi-cloud.

Yes. What did it do? It made Security Command Center, both its left hand of posture, misconfig, vulnerability, and its right hand of threat detection apply not just to GCP, but AWS and soon Azure as well.

Yes.

And it did that by bringing in Chronicle security operations as part of your CNAPP experience. And so with apologies to Neil MacDonald, I think what we're witnessing is the birth of Gen 3 CNAPP.

I'm cringing here. I'm sorry.

It's my first day as an analyst.

It's your first day as an analyst, but you know, you have to kind of have at least Gen 1.5 first before you really go to Gen 2.

Well, Gen 1 was Prisma, which is just a bunch of stuff glued together. Gen 2 is Wiz, and Gen 3 is SCC Enterprise because it comes with a security operations suite as a core part of the offering.

I think that you should look for a job in marketing, but then again, wait a second, PMs pay more, so maybe not.

That's right. No, I'm going to stay where I am. Thank you.

Other launches.

Okay, other launches. What else have we got?

I think that applied threat intel, it actually sounds kind of like basic and obvious. Sure, we have threat intel. Sure, we apply it. But ultimately, my ex-analyst life taught me that a lot of people get threat intel and then they sort of like don't know how to profit.

Well, we've talked about where they read it.

Yeah, or they do it for entertainment.

Yes.

Yeah. So the tricky part is that using it in a product, everybody says, yeah, sure, I have feeds. They go into my SIEM. There are detections. But there's a lot more work that's needed for it to actually be real.

Yes.

And I feel like our implementation of Applied Thread Intel actually makes Intel come to life and deliver value. And yes, now I know I sound like marketing now, but it's also very clear—

Anton, I encourage you to get a marketing job, except OCISO presumably pays better.

Yeah, there's that. But it's the—

Apologies to the marketing people we know don't listen to this show.

Who are not listening, yes. That's why it's pretty safe. The point is that it's genuinely more effective than just, here's a feed, stick it into my SIEM.

Yes.

And a lot of engineering work, not marketing, went into actually shaping it to do that.

Yes. Yes.

And then I think Chrome did something quite amazing.

Yeah. What did Chrome do?

Yeah, I guess the incorporation of all this zero trust stuff right into the browser. I mean, their naming, probably not my favorite because now there's Enterprise Premium, Enterprise Premium, Enterprise Premium. Yeah, exactly. But you had a product called Security Command Center Premium, so it's kind of partially your fault.

Which we now renamed to Command Center Enterprise.

And now the next version will be Enterprise Premium, so it's kind of like—

No, no, no, no, no, no. The next version will be Enterprise Plus, thank you very much.

That's not better.

Well, listeners, it's hard to name things. My parents struggled so much with naming, they named me after a dog. Fun fact.

So kind of like Indiana Jones.

Yeah.

Okay, but that's an epic— that's epic DNA to have in your name.

I'll take it. I'll take it. That's the nicest thing anybody's ever said about that. So, so we had a bunch of launches. We had some sessions. You and I each had a session. We talked about your session this morning. I had my session in between then and now. That was fun. It was my 5th SCC session at NEXT.

Not on this NEXT, across all the NEXTs, right?

All the NEXTs. All the previous nexts, which is a fun phrase. But the good news is I didn't have to do the demo today, listeners. So I slept like a baby. I woke up feeling serene and I'm a happy camper because the demo went great and I had nothing to do with it.

By magic.

By magic.

I'm pretty sure a little LLM knows.

By my colleague Jason doing a great job. Thank you, Jason, wherever you are.

Thank you.

So sessions were good.

I think that our guest sessions, right? We did the session from our Tuesday guest, right? Payal was about SecOps and AI. And I think we, I wouldn't say grilled, but we got some really good responses from her based on the use cases. I think we're kind of slowly moving to reflecting on our podcast episodes, right? Because we had, this is the 4th. So we did one on SecOps, one on AI, and one on cloud security, which is coincidentally how I spend most of my time, these 3 things.

Me too. Yeah.

That's weird. Like I bet we were a little biased maybe, but like we reinforced each other's bias in picking these sessions. We didn't have anything on HSMs.

No. Are those interesting?

No.

Listeners, if you wanna hear about HSMs, let us know in the comments.

We are now silent because we are waiting for those comments to immediately show up because—

Which are not gonna show up.

No, they're not gonna show up. There was also a session which we haven't attended on SecLM. They actually did run it and I've heard some good news about it.

I heard the demo was excellent. Yes. Yes. And I think, Listeners, if you're curious about SecLM, stay tuned. We'll have more content about that soon.

Yes. Yeah.

I've seen some interesting stuff on the floor.

Oh yes. We were doing our normal kind of like our standard wandering the floor of the event, which we do usually at RSA. Yes, correct.

This is our first time doing it at Next.

Didn't you not do it last year?

The Next of previous years, past Next, past Next was too small. This year, It's enormous.

It's enormous. And moreover, there are enough security vendors to actually do a security vendor walk. Yes, there's some new school security vendors. Actually, as an ex-analyst, I categorize it into 3 buckets. There are new vendors, there are new new vendors, and there are old vendors.

What's new new? You're not so good at naming either.

The new is basically the cloud era vendors. Some of them we mentioned, all the Orcas and, and similar ones that are clearly new compared to the firewall makers of the 1980s, maybe '90s. I don't know. It doesn't matter in this case. But there are also vendors like the one we talked about, the runtime security. They're cloud security, but they're of the later generation.

Upwind.

Upwind, correct.

Yes. Yeah. We met a vendor here called Upwind. This is not a plug for them, but we met a vendor called Upwind that's only 18 months old and they're already doing or saying they're doing all the things in CNAPP 2.0.

Oh, someone might say CNAPP is square. quite large. And again, I had a chance to be reminded of that by the acronym creator himself, Neil McDonald, who I met at the analyst meeting.

Secret society.

No, no, no, nothing secret. Yeah, it was an analyst secret society. That's right.

Yeah, listeners, I didn't have the right handshake to get in the door.

He reminded me of the CNAPP, and then there are CNAPP vendors with different lineage, right?

Yes.

Some new, some old. Snyk is a good example, right?

Yes.

Their lineage is clearly in application security, doing things To the left, but they're building things that are more of the detection response and things that from the right side.

Yes.

So the CNAPP may actually happen. Am I really saying this?

I don't know. I think you have a world where there's a mountain and we're all John Henry-ing from different sides.

Yes, digging tunnels, and in the middle there would be like the actual CNAPP.

Yeah, yeah. Eventually we will unearth true CNAPP in the center of this mountain.

That's profound. Now I need to like go and reflect on that. But it's actually kind of happening and it gave more meat to the concept. And I think that Neal has a history specifically of inventing something and have it materialize 10 years later. Security is becoming a big data problem is his paper from 2009 or '10.

No way.

And then all the data leaking stuff started happening roughly in 2015, '16.

And now it's Neal's Chronicle.

Yeah, right. Actually, yes, 2019 Chronicle. But the paper that started this whole data leaking, data, big data for security that he wrote was, I think, 2010, if not 2009.

That's amazing.

So CNAPP may absolutely show up at your door, just not immediately.

Best way to predict the future is to invent it.

Correct. And live long enough for it to see fruition. That's actually kind of important because some predictions are happening, just not at the timeframe. You predict the thing, but not the timeframe.

Yeah, that part's important.

The other CNAPP, remember this LLM demo where LLMs are like competing in the game of cloud remediation? That was quite epic.

So we stopped by a vendor's booth. Do we want to name the vendor?

I think we can name the vendor.

It was Orca. We've had them on the show before. We saw a remediation demo where they used 2 different publicly available LLMs to suggest remediation for the same issue. And we looked at them side by side. And I don't just say this as a Google employee and shareholder. I think the Gemini one was better.

I think the Gemini one was better. But it's also interesting that if the LLM bots learn from the same docs and the same practices— They should come up with similar things. It is unlikely that they would make a basic mistake with these things, which are precise. They may make basic mistakes in other things.

Grounding.

So grounding here was very in evidence. They're probably both grounded by our docs, by Google Cloud docs.

By their own docs. Yeah.

But you caught something that was interesting because there are probably different ways to achieve the same problem. You can undo old settings and then reinforce new settings, or you may do something else that will achieve the same outcome. So, the other interesting part connected to this was that when we sort of jokingly prodded the person in the booth, do you want to just run this code? And that was about Terraform. The person was mildly offended and said, no, you want to give it to a human because Ultimately, this is really cool and amazing, and it actually can be often very accurate, but is it blindly trusted to just be pasted in the prod systems? Probably the answer is no. No, the answer is for sure no.

The answer is for sure no.

For sure no.

And so, there's still a big trust gap between what's coming out of LLMs and what comes out of, say, your build system.

I think that's a bit harsh. There's—

No, you were saying—

There's a big gap. But I don't think it's a big gap.

I think it's a pretty big meaningful gap. I don't look at the output of a compiler and think, I better look at what it compiled before I run it.

No.

I don't look at the product of a build system and think, I better double-check the build system's work before I run it.

No.

There's a fundamental categorical difference between that and LLMs.

Yes. Yes. Yes. In that sense, there's a big qualitative gap between the two. Yes. That part is interesting.

And I think that When does that go away? What's the prediction on when that goes away?

You know, I've met 2 camps to this extent. Like, there's a camp that I call Muse, not Oracle, right? There's an article in the early days that basically said LLM is amazing, but it's a muse, not an oracle. You're not going to get the right answer for it probably ever, but you're going to get a very useful answer that did help you along the way. And I think I am in the Muse, not Oracle camp for many things. And it's very useful for me, like, for writing, for research, for other things, for shaping ideas. But there are enough people who want an oracle. And the remediation recommendations, you don't want a muse. Let me inspire you to write the best possible remediation Terraform. No, you want the actual Terraform to get written and to paste it somewhere.

Now, does that get to kind of the part of the other conversation we're having about soul? There's no soul in Terraform.

So, this can be really fun. And I think that it's—

Listeners, this sounds like a wild goose chase.

It's really not because it was about which tasks we would successfully give to the robots, to the machines, and have them execute it to the best of our requirements and needs. But which tasks, like compose great music to inspire me on my down day, people still debate this. I think there are camps that say that robots would do it and there are camps that say robots would not do it. Here briefly means machines, not actual robots. We saw a robot dog, by the way. We saw a robot dog.

I had never seen one of those Boston Dynamics dogs, listeners. I knew I would be afraid of it when I saw it in person. It was even scarier than I thought.

And my case was from the pictures, it looked kind of cute and a little unwieldy, but like from reality, it was actually more scary. I would go with that.

It's a very capable object.

It would probably stop me making Robot Rebellion jokes for at least 2 weeks.

Yeah, it's scary the same way like a snowblower or a lawnmower is scary.

Lawnmower is not scary.

If it runs you over. Have you not watched Mad Men?

Yeah, okay, fine, now it's scary.

Guy walks into an ad agency.

Yes, so it's a little scary from that point of view. But yes, like, it's a long detour from our topic, but I think that if some of the paper attacks and some of the like digital IT stuff starts affecting physical world, Think about the jailbreak on the robot dog.

You could use it to break out of jail.

Oh, cool. Jailbreak to have robot dog break you out of jail. That's profound. No, it's not. I think we've been on the show floor a long time, listeners.

We think things are profound. They're not profound.

Yes, I exaggerate.

Yes, but definitely food for thought.

Food for thought indeed. What about the cloud island? That's also food for thought. I think we are down to fun ideas we had while doing Next.

Yes. So, Cloud Island. Is cloud security possible to safely treat as an island? Or should cloud security be a peninsula? And have we seen historic examples of other security islands, which themselves were tectonically merged into the bulk of the mainland?

And to try to simplify and summarize what Tim said.

What the fuck did Tim just say?

Yeah. Well, to try to summarize it, I didn't tell him I'll do an LLM on your speech. Like, a lot of people kind of— some people hinted to us that maybe cloud security isn't a thing. And we were like, what do you mean not a thing? We have a whole podcast called Cloud Security Podcast. It's clearly a thing. And they said, no, it's not a thing because if As many Indian people pointed out in our podcast here, if your on-premise Active Directory is hacked, your cloud is hosed. So, your cloud security is—

Not an island.

Essentially hijacked by a broken on-premise component. But then, have you failed at cloud security? If it were an island, you can't fail—

Over there and suffer here.

Have it be affected over here. But if you Often fail with Active Directory, always fail. Your cloud is subjected to a lot of torture as a result. So, if cloud is truly an island, that wouldn't be happening. So, they kind of reminded us that maybe a lot of cloud security things aren't cloud security. It's just security that happens to include cloud. Can we rename the podcast to Security Podcast Including Cloud in Many Cases by Google?

Absolutely. That would get through naming committee because it's accurate and descriptive.

Yes. If you convince them cloud security is not a thing, then the naming team would withdraw the approval of the Cloud Security Podcast name, and we have to be named Security Podcast, Sometimes Including Cloud by Google.

That's not very enticing. I think the other sort of interesting question that pops out of that though is, if you look historically at land masses which were isolated, And then through sea level change or tectonic action, joined up with land masses otherwise, bad things happen to the native flora and fauna.

Of the island, not of the continent. Of the island.

What happens to the cloud security flora and fauna as it merges back into the mainland?

I can tell you one failure scenario, which is the reasons why initially I argued against cloud security is not a thing. It's just because Some people who believe that it's not a separate thing sometimes are the people who want to go back to the '90s data centers and basically run cloudsec as they run netsec, endpoint sec. They want to run agents everywhere. They want to do firewalls. They want to do everything they've done in the past because they say, wait a second, you convinced me cloud security is not the thing, then there's nothing new, then I'll just do what I did 20 years ago. And that's the failure mode.

Is our beautiful era of agentless security going to come to an end?

No, because I think that's a corrupted version of cloud is not a thing.

Oh, that's the wrong take.

That's the wrong take. But that take doesn't invalidate the position that maybe cloud security is going to merge in the main body, kind of like mobile security did.

Yes. Oh, you buy me on mobile now?

Yes. But that's like we're looking at the past and now future, what would happen with AI security?

Right. And my theory, listeners, is that AI security is like the further out piece of tectonic plate pushing cloud security back into the mainland.

Oh, so there are 2 islands. One of them is close to shore, it's CloudSec, and then the other one is coming in behind from shore, pushing us in. I actually see that almost.

Yeah, because AI security is requiring you to rethink cloud security and focus, as we discussed with Sean, on the same foundations in cloud as we have on-prem.

So you're implying that nobody trains LLMs on mainframes?

I would bet it's hard to find somebody who trains LLMs on mainframes. Yes.

Yes. Not to assume that everything pre-cloud is similar to 1970s, '60s, '50s mainframes. '50s. Okay, I'll stop at '50s. But the idea is that, yes, most of the AI security implies that some components run in the cloud.

Yes.

Likely at least infrastructure, probably applications, and probably also models.

Yes.

So, in that sense, yes, the island may be pushing the other island closer to shore. What kind of situation, what would arise when cloud security would start— would stop being its own thing, but still would start being reabsorbed and not through a failure mode of on-prem takes over? Like, cloud would affect the team composition, would affect how people actually organize things. It would affect the practice of network security.

It would affect application security, supply chain security.

Supply chain, data security for sure, because data security would probably morph less, right? Like it's more—

Yeah, the fundamentals there seem pretty fixed.

Sure. Like in today's AI talk we did with Sean, our guest and also my co-speaker, data governance came up and I kind of said, I think I even did this whole data governance, I'm going to go to sleep.

But now you better wake up about it.

But now I have to wake up about it and I have to get excited about it because it spreads its tentacles all over AI security. And you can't do AI security without data governance, without data security. And a lot of stuff that sounded very old and stale and boring becomes like, holy crap, if you don't do this, you're so done.

Yeah, yeah, yeah. So that's a really interesting example of where this AI is pushing cloud into the mainland and a mainland thing is coming in very fast. That's a great example.

Yes. And I think mainland would modernize at least somewhat, right? Like, yes, like I've seen people even the whole looking at the domain of DNR, like we talk about modern SOC and how SOC may not be a thing if you have engineering-led mentality. It's like an obviously obsessive focus of mine, but it's, it's not happening very quickly.

And Google.

Yeah, and Google. It's not happening at high speed.

No.

But I think the more cloudy, cloudy concepts, cloudy models, cloudy frameworks come in, the more that would be sort of seen as normal. But it may, as with Neil's case, it may take— if you think 3 years, it would then be 7 years. And if we think 7 years, it may then be 10 years. But it's probably going to be happening.

But it's coming.

It's coming. And that's— yeah.

Yeah, that's really— that's an interesting observation.

It is. By the way, do we have to ask each other questions about reading materials and the resources? This is a favorite closing question, and we may have to ask each other that question. So this is kind of fun, and I did not prepare it. I just thought about it. Sorry, Tim.

Okay. Well, in that case, I'll go first. Anton, what do you recommend people read?

Okay, but I don't get to drink if I say my blog. So you torture everybody else by saying, don't recommend Anton's blog, everybody does this. But I can do this because it's my blog and I own the freaking thing. So I would say lately I kind of mix the SecOps and AI security a little bit. I'm a lot more careful with AI security because of the insensitive topic and all that. But ultimately a lot of the transformation changes affecting security by cloud, by other modernization efforts. It's kind of fun, not only SOC. So to me, I'll just point you at some of the detection engineering posts, but okay, fine. I'll do something else. Not just that. Okay. Now I'm going to suffer because that's when I did think about it immediately. Okay. Actually, I have some peculiar piece of reading. Strangely enough, I've been trying to learn more about IAM. And remember there was an episode on our show.

It's faster to reset.

Yes. Remember that episode on our show when I made fun of my former Gartner colleague who used to run an IAM team at Gartner? I said, but Ian, it's all password changes. There's nothing else. And of course, he stared daggers at me and we had a friendly kind of resolution of that. But I'm kind of, for some strange reason, this IAM is coming to haunt me. And I've been reading up a lot on how various cloud providers do IAM.

Yeah.

Both our materials, others, what Gartner has written. So, I've been kind of reading up on IAM and why it pains people so much. And it does in cloud.

Interesting.

Oh, and one tip.

One tip.

Okay. You know what? I'm going to steal the tip from Sean, but I'm going to modify it a little bit. And a lot of people sometimes just give this throwaway tip of, oh, you need to focus on the basics. Don't buy this shiny—

No, we don't say basics anymore.

Right, exactly. But I'm getting to this. So, and that tip is a very throwaway and sometimes even kind of like, Exclusive, not inclusive, because it implies that it's just easy, you're just lazy or something. So, it's a very judgmental way to do things. Hey, just do the basics. And even saying the version that Sean took is do the fundamentals first is good, but also it's not very specific. But I think that I would turn this into the following. If you have a choice of picking a random, cool, exciting piece of new tech that promises a new way for solving a problem.

Yeah.

And you can dedicate time and money to this. or you have a known piece of more traditional security advice that you sort of need to know, need to do, but haven't, my tip would be try to intelligently pick the second and not the first. I'm not trying to be anti-capitalist, anti-startup. I'm just trying to double down on the fundamentals. And sometimes, as I think one of us said at the previous podcast, it's not about fixing the foundations. It's not even fixing the wall. Some of the startups is about the new paint for the roof. You may not need that. You do need the solid building. So supply chain, a few other things that are more of the bottom of the stack stuff better get attention. I don't think we can roof paint our way to security. I think we have to do the basics. Oh, I said the basics. Fine. Fundamentals, foundational things.

Yes.

So, one, Tim, favorite reading? And one tip, and of course I have to warn you, don't offer my blog as reading.

And listeners, this is funny to hear Anton do this, not only because he's asking me, but because it's almost always my job to ask that question. I will say recommended reading, I'm gonna go outta left field. I'm gonna recommend The Left Hand of Darkness by Ursula Le Guin. Oh, Ursula Le Guin. Science fiction for everybody, but really, really wonderful science fiction. Hmm.

I like some of her books. I don't think I read this one. So at least for me, I'm gonna, it's gonna go on my to-do list.

Very good. Very good. And then a tip. It's funny, you actually stole the tip I was going to give. Foundations.

I'll give you a version of that tip. It's fine.

No, no, no, no, no. It's okay. I think what I want to give people the advice is to stay curious and not think that you actually have the answers to this stuff. Because what I've learned in the last 6 years at Google is this stuff changes in ways that you wouldn't imagine. And I wouldn't have guessed that AI emerging would push cloud closer into traditional stuff. That insight we've had today, deeply surprising to me. And if I was not staying curious in this role, I never would have seen it.

And it's also genuinely, it has this unique combination of not obvious right before and sort of ringing true right after, which I think all the best insights are.

Yes. So with that, listeners, thank you for joining us for a couple of next live streams. We're going to do our best to get these turned into proper episodes that we put into the channel. Worst case, you'll find them on YouTube. Best case, you'll find them in your podcasting app Monday morning, piping hot because you hit that subscribe button. 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_Chuvaikian 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.