EP163 Cloud Security Megatrends: Myths, Realities, Contentious Debates and Of Course AI

Security running.

Hi there.

Welcome to the Cloud Security Podcast by Google. Thanks for joining us today. Your hosts here are, as ever, myself, Tim Peacock, the Senior PM for Cloud 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.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, the rest of our listeners, all on our community page. And if you haven't checked out the Cloud Security community pages, find the link in our show notes. Anton, this is a fun episode because we have our Most frequent flyer returning to us once again, Phil Venables, talking about megatrends. And as I understand it, since I wasn't part of the recording— listeners, you get an episode without me— we added one more trend. What could it be?

Yeah, well, given that this is what, 2024?

It is.

You can probably guess what the megatrend, like what's big and important and everybody cares about this. And no, it's not cloud, even though it is a cloud security podcast.

It's not cloud.

It's not cloud. So it's not—

Is it security?

No, security is too 1990s.

Oh, we already care about that. Yeah. It's not podcasting.

It's not podcasting. I hear podcasting is in the, like, moving to the, what do they call it, in the hype cycle, kind of moving in the trough of disillusionment.

Trough. Yeah, yeah, yeah. So it must be AI.

Exactly.

Yes.

Thank you. We have a winner. We have a winner. Excellent. It's AI.

And what do we have to say about AI? I guess we should listen to Phil about that and maybe Without further ado, let's turn things over to Phil Venables.

Hello, and welcome, Phil, to the podcast for the 8th time, actually. We had to count. This is amazing.

Yeah, always great to be here. Glad to be a regular guest. So it's all good.

Perfect. So we wanted to revisit the megatrends idea. And I remain a huge fan. It sort of opened my mind when I saw the original blog of 8 Megatrends. And it's hard to believe it wasn't 2021, and this is what year is this?

2024.

So, where are we now with them? I mean, I know we added one more, but what's your overall assessment of how the idea fared?

Yeah, they've definitely stood the test of time. And so, you know, for people listening that don't quite remember this, you know, back in '21, as Anton said, we identified these 8 megatrends that were kind of driving cloud and cloud security in particular. And I won't kind of recap them. We'll put the original blog post in the show notes, but I'll just, I'll pick a few and give you an example of where we think they're being kind of sustained or maybe amplified. So, one of them was something we called the economy of scale. And, you know, this was the notion that we can use the hyperscale nature of the cloud and, of course, especially Google Cloud to keep raising the baseline of security by reducing the unit cost of control. And this is proving more and more true. I mean, we continue to invest massive amounts, almost like disproportionately high amounts in security, not just because it's self-evidently important for us and our customers, but also because we can amortize that cost across this huge fleet of infrastructure that we have. So, if you spend a few hundred million dollars on something to improve, radically improve security, and you amortize that across millions of servers, the cost per server is not that much compared to what you might do if you've just got 10 servers on-premise. So, this kind of keeps playing out. The second one that we see see a lot, and it's definitely resonating more, is this notion we call shared fate, which is our means of saying we need to move beyond that line of shared responsibility between the cloud provider and the customers and do a better job of protecting customers, not just on the cloud, but how they configure themselves in the cloud by giving them better guidance, better tooling, better solutions, and also most importantly, better secure default choices and more proactively secured configurations. And I think, you know, where I wouldn't claim we're perfect or any other providers are perfect yet, but we're making significant strides there to make things ship with full safeties on so that people have more control over these things. And you see many government agencies, you know, particularly DHS, have been putting out a lot of correct and great messaging on the need for secure by default and secure by design from the big tech companies, which I think is great. And the third and final one I'll mention is this notion of cloud as a digital immune system. And so this is proving out even more since we first wrote that, the original post, because since then we acquired Mandiant, and we've got an even bigger collection of threat intelligence to augment our already massive threat intelligence that we had. And so from that, we can see attack activity incidents, and almost in real time, in some cases, feed that into our systems to proactively protect ourselves and our customers even before those attackers may have had time to come to them. And this notion of real-time or near-real-time adjustment in response to what we see at a global level, it plays out as a kind of an immune system. And I think that's definitely— the flywheel of that is happening more and more.

And I think Mandiant is a really good addition to that specific bucket. I agree, because ultimately, More intel is good. Better intel is good. But the best intel in the world is better.

No, that's right. That's right. Yeah. And just like the different types of intelligence that Mandiant brought us compared to what we picked up historically through the wider Google sensory apparatus. Yeah.

That actually makes sense. So we added one more. So now it's 9 megatrends. And I think that at the risk of sounding slightly buzzwordy, the 9th is, of course, AI. So what made you add it? Like, what made you think, hey, AI really is an equal partner to the other 8 megatrends?

Yeah, I mean, it's interesting because the other way of looking at that question is like, how could we have not added it in 2021? And so, you know, what were we thinking? No, but in reality, we did, you know, we talk about in the update to the blog post about AI being what we call the sleeper megatrend. And the reason we kind of use that phrase is back when we first looked at the megatrends, we did actually think about AI. But we thought it was so embedded across all the other 8 megatrends that we didn't think to call it out. And the backdrop to this is we, especially at Google, but not just Google, have been using AI, particularly deep learning-based AI, for well over a decade to defend billions of people. I mean, we use it to defend Gmail users, Chrome users, cloud users, filtering malware, identifying malicious websites, identifying threats in the cloud, detecting misconfigurations of privilege. It's just everywhere. And so, So it felt like it was just this common underpinning of everything. But over the past couple of years, as everybody has observed, the rise of generative AI as distinct from more traditional deep learning predictive AI, it provides a whole different set of capabilities. And it's not just going to augment the other megatrends, it transforms them and other things dramatically. And so it felt like it was right for it to be a megatrend on its own just because We're already seeing it being so transformational for security.

But one thing that we— this brief debate at one point about the 9th megatrend, and I felt like these are cloud security megatrends. So when we think about why AI is specifically a cloud megatrend, to me, it was very logical and obvious that if you're doing any kind of innovation or work with AI, you're doing it in the cloud. You're not buying appliances or servers or hardware. But I've met people who sort of like, agree with the AI impact, but they're debating the cloud connection. Could you sort of reinforce this for some of the listeners? Because I got convinced of that, but it took a little bit of arguing. Like, well, what if I do AI security on-prem? Then am I partaking in the megatrend or am I not?

Yeah, well, look, I mean, I think they could well be. I mean, you know, these are cloud security megatrends and cloud megatrends, but, you know, most of them are actually also megatrends in general. You know, one of the other megatrends was software deployment velocity. Another one was software-defined infrastructure. They can work on-premise just as well as in the cloud, although it's a little bit harder because cloud was designed to do that and on-premise wasn't. You're kind of retrofitting that on-premise where you get it for free in the cloud. And I think when you look at AI, you not just need the processing power of the cloud, you probably also need the data governance, the control, the security to manage how you're developing, training, and using AI. And also a big part And this may be a statement of the obvious, a big part of AI is marrying that with data, both for training, testing, and ultimately for serving up and using AI. And most organizations, when they think about what they want to do with massive-scale data processing and analytics, they're coming to the cloud for that simply because of the scaling properties of the cloud and the tools around data that exist in the cloud. So, If most of the data is already in the cloud, then deploying AI in the cloud makes sense rather than bringing it all into the on-premise environment. And I think just the interplay between the tooling of data, AI, analytics, storage, processing power, and the fact, especially for Google, unlike others, is we have our own AI technology that we, as well as partnerships with other companies that we bring to the cloud. So it's just easier to do in the cloud, I think.

And I think that what you just presented also explains this argument you made in the previous episode we did focused on AI, where you kind of said that AI, in your mind, medium to long term, would favor defenders more because defenders have the data.

Yep.

And that to me is really a line that's, I mean, maybe not like transformed my mind, but it's like it actually opened my mind and made me very convinced that, hey, if I'm going to go argue with somebody at a conference and they say, AI attackers, hackers would do these bad things, I'd be like, Hold on, who has more data to eventually train the AI? So, to me, that's a really good argument that I became enlightened after our conversation specifically.

And actually, since we spoke about that, I kind of think about that more as well. And it's interesting. I don't think it's— it's not just the data. I mean, this is kind of a metadata property, but it's the context of your organization as well. Because when you're looking for AI to be trained about how to manage configurations and detect misconfigurations, as well as all the more traditional security work. You not only have your own organization data that learns how you want things to be configured, but it understands the broader context of your organization, so it can give you more meaningful insights in a more timely way. And attackers just, you know, unless they're completely owning your environment and training an attacker AI just on everything about your environment, Then they're not going to have that advantage. And clearly, if they're already owning your environment to that extent, they probably don't need AI to try and break into it.

Yeah, that makes sense. And I think that it's also interesting. I don't want to turn this into another AI episode, even though I'm super tempted to. But one thing that I keep hearing in various conversations is the whole governance question. And people who think governance is maybe dry and boring get excited about this, and they fear this type of use of AI by business, use of AI for attacks, or just abuse of AI. So good governance seems to be at the center of all this, using AI securely and also using AI for security. So do you have any secret recipes for CISOs on success with AI governance? And please don't tell me build a broader team, write good documents. Like, everybody sort of knows this, whether they do it or not, separate story. But like, is there anything like— profound in doing AI governance right?

Yeah. Well, like, the easy answer is go read our Secure AI Framework documentation that we've published. And, you know, we're expanding all of that. Yeah, we'll put that in the notes as well. But there's a lot of good stuff there. But I would say when you strip all of this governance and safety and security and trust and all these frameworks and all these thousands of pages in dozens of documents that are coming out from every vendor and every think tank you can think of, you know, for me, when you strip it all back, it's It's just 3 things for CISOs to pay attention to. The first one is this is a data problem. So you've got to protect the training data. You've got to protect the test data because the test data can have untoward feedback into the models if the test data is attacked. You've got to manage the end-to-end integrity of how data gets to the model as part of using the model. And so at the end, fundamentally, this is a data governance problem.

Yeah.

And focusing in on that helps solve probably a big chunk of what you might otherwise describe as AI governance. The second thing is how you're deploying and using the AI. So, again, when you're deploying an AI, it's being deployed for some purpose, it's getting some input, and it's triggering some activity or providing a response. Well, you've got to manage what gets to the AI, you've got to manage how it responds. If it's actuating some physical process or creating some data change or transaction change, you've got to put it through some kind of guards and circuit breakers and other things. And so, there's governance of how do you wrap it in some layer that controls the behavior of the AI, to be honest, just like you would do with some conventional software in critical environments. And then the third and final piece is how you interact with other teams in your organization, whether it's compliance or legal or you know, the data governance team or any of these other teams. So really, it's just that. Manage the data, manage the operating risk of the deployment, and then find a way to partner with all your other control and risk teams who are probably developing their ways of doing it. And if you get that right, you know, everything's probably gonna be okay.

I like how this is comprehensive, very frameworky. My ex-analyst brain really loves it. And it's also not like truly magical, as is like Phil wraps up a magic wand and says, just wave this 4 or 5 times and you have it. But I think that many problems in security don't have that type of answer, like whether AI involved or not. So to briefly pursue this topic for one more inch, I guess, is, so do you hear any interesting questions from CISOs on this? Or maybe more importantly, are you not hearing any questions that they should be asking on the topic of AI and security?

There's a lot of focus from CISOs at the moment, I think, on how they prevent their organization's information leaking out into some of these public chatbots. And so there's a lot of talk about AI control as being this yet another perimeter blocking thing, largely because some of these external chatbots, you know, have had a reported tendency to leak queries back into the foundation model, which gets served up to somebody else.

Yep.

I mean, I would say, you know, we've taken a lot of steps in and around our environment to reduce the risk of that and to manage all of that type of control. But I see a lot of security teams focused on that and potentially not paying enough attention to the fact that all of their business units and IT teams are buying AI products and taking the AI and integrating it into their business processes. And so to be clear, I think there are a lot of security teams who are all over that and managing it well and managing the governance and taking advantage of it, especially as they look to use AI for themselves. But I think if you look at a broad brush out there, There's a bit too much focus on treating this as a block access to external chatbot website and not enough on what are my business and my IT teams doing in bringing in products and integrating AI into my business flows.

That actually makes sense. And I think I'm also a little afraid when people start talking about blocking the— their approach to AI security is block the website. I'm like, No, if that's how you're doing it, not going to be good.

In the meantime, one of their main applications has got an API interface to some product.

The same exact product they're blocking on the website. Yeah. Okay. So this was hugely useful. And I feel like I want to go back to megatrends and specifically to the maybe contentious, like we had 8, we now have 9. And I guess one of them feels like it's more contentious. And that's not a leading question. Sorry, Phil. Which one is the most contentious?

I know, we all get, you know, you and I and everybody gets the same grief. And it's the one megatrend that everybody throws rocks at us on is this megatrend of cloud driving simplicity. And, you know, interestingly, you know, we stand behind that. And I'll kind of, you know, some of you have heard me talk about this before, but I'll kind of repeat the analogy a little bit is, if any of you have ever worked in like significant on-premise environments, those are massively complex environments. It's just that very few individuals are ever exposed to the complexity. So, any reasonably sized system in any large corporation that does any type of transaction processing, kind of customer ordering, financial systems, industrial-scale management, whatever, is typically made up of, you know, hundreds of servers, multiple different operating systems, multiple databases, multiple transaction brokers, cluster management, multiple storage systems, networks. I mean, it's just a huge amount of complexity. But because you've got storage admins, network admins, operations teams, software developers, security engineers, software architects, you've got all of these people, no one person's ever exposed to it. So, this complexity just has historically just, like, grown under the environment.

Yep.

Now, when you come to the cloud, the nature of cloud, again, being a software-defined environment, you just see everything at the console. And you as a developer or a DevOps person or a cloud engineer, for the first time, maybe, you know, you're seeing the ability to manage storage configuration, network configuration, and all of that's together. Now, it's all massively simple. I mean, if you go into any of the cloud storage console, you know, the console and manage cloud storage, that is way, way simpler than sitting down at a storage area network controller or a storage array, you know, console interface, if anybody has ever done that, you know. You know, if you haven't, go back and look at some of that documentation, then you'll realize just how simple the cloud is. But it's great that people kind of push back on this simplicity notion because it does drive us. And this is essentially the megatrend, which is, the megatrend is not it's simple now. The megatrend is it's always getting simpler because we expose the complexity, which causes the feedback loop to cause us to make it simple. And again, the classic example is Kubernetes. Everybody criticizes Kubernetes for complexity. But, you know, again, anybody that's ever built a high-performance cluster management transaction processing distributed application system knows that that's complex. Kubernetes is not complex compared to that. But the fact that people perceive Kubernetes as complex has caused us to put abstractions in, which makes it easier and easier to manage and deploy. And we have things like GKE Autopilot now, which is very easy to manage, you know, in relative terms. But the trend of people complaining about complexity is gonna keep making things simpler and simpler. And I think that proves out the megatrend.

Your storage example is actually even worse because ultimately complexity starts with, I need to buy a server or a storage array. Where do I put it? Who owns that part? Do I have electricity? Do I have, like, the complexity goes to start at buying a physical box and goes all the way to cluster management. So, the actual chain of complicated steps is much larger. And you say, oh, this console is Well, try looking at the procurement process.

No, no, exactly. And then, you know, you take it in another direction and say, you know, I've worked in environments where I've had to deploy encryption across storage, across multiple different storage array networks, multiple different types of media types and disk arrays. And when I sit down in cloud, I don't even— well, at least for Google Cloud, I don't have to tell it to encrypt the storage. It's just all encrypted by default. And I never even have to think about that complexity. And that's an extreme example of simplicity having also delivered on security.

That makes sense. And I think that maybe this episode would help people who are kind of still pushing hard at that. So, the other one I wanted to touch very briefly is this whole software-defined nature of cloud. I don't wanna make any server hugger jokes, but I still see people who are kind of stuck in the cloud is just somebody else's computer mode, which completely neglects the software-defined nature of cloud and, of course, of cloud security. So do you think in your mind many enterprises are— users are growing to appreciate the software-defined nature of cloud and cloud security, or is it still a bit of a sticking point? They want their, you know, servers and software.

Yeah, no, I think they do. I mean, I think this one is an interesting megatrend because this was the one that, you know, one of the main ones like software deployment velocity that didn't just apply to the cloud because there's obviously plenty of organizations that have deployed on-premise cloud-like environments whose main characteristic is that they're software-defined. I think the difference between the on-premise and cloud environments is you don't really have to find the edges of what's software-defined because it's all software-defined in the cloud. In an on-premise environment, you've always got, at least the ones I've built and worked in, is you've got some of your, for example, an on-premise Kubernetes cluster is all software-defined. You've got some other types of— VM configuration that's software-defined. But then it, going back to our storage example, it then interfaces to a storage array network with a whole bunch of edge layer code to bridge the fact that that's not particularly software-defined, or at least not software-definable in the way you software-define your on-prem cloud-like environment. So, I think it just becomes easier in the cloud because it's more pervasively software-defined. Yes.

And I think that it's an acquired taste for server admins and the classic IT leaders. At least that's my impression. Maybe the description is acquired taste is something they would love, but they do need to go through some kind of a pushing through process to love it. So, on the kind of more exciting positive side, like maybe I should use my 3 AM test. If somebody wakes you up at 3 AM and says, Phil, which megatrend is like the most strongly manifested? Which one is like you see it everywhere? So, what would your answer be?

I think it's a close call with economy of scale because that's almost like so pervasive that you can't even, maybe even don't even think of it as a megatrend anymore. It's maybe just so obvious that why even call it a megatrend? But the one in addition to the economy of scale one, I think, is that whole notion of cloud as the digital immune system, because that just keeps playing out. And that's an interesting one. That's one that's even more especially amplified with AI and what we're deploying on some of our AI tooling, where we've trained large language models on our threat intelligence and other security data. for that to be in the loop managed by humans to drive that feedback loop of taking threat intel to detections and configuration fixes. So I think that's just going to keep— the flywheel of that is just getting faster and faster all the time.

Yeah. And I think that I didn't want to answer this question myself, but actually, the number one is probably the unbeatable one because you can never beat the economies of scale in one environment. So that probably proves that cloud security would always— is almost like doomed or destined to be better than on-prem security because we can make use of economies of scale. And you have to be really, really large and you'd still be smaller. So that actually makes sense.

No, I think that's right as well. And as we know, many of these megatrends interact with each other to also be reinforcing. So you've got these kind of flywheels that go on inside each megatrend, and then you've got an even bigger flywheel where all of the megatrends kind of feed each other, especially now we've put the 9th megatrend of AI in there as well.

Perfect. This was super insightful as always. Really appreciate your time. And again, this is probably one of the best episodes we've done.

No, always happy to be here. Looking forward to future discussions.

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 @anton_chuvakin and _timpego. 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.