Cloud Security at a Breaking Point: AI, Complexity, and the Future of Trust

Hi, I'm Joan Goodchild and welcome to another episode of Cyber Sessions where we talk with security executives about the trends and topics that are influencing the cybersecurity space. Today we are at the CSO Cybersecurity Awards and Conference here in Nashville, Tennessee.

And I'm really excited to be joined by Jim Reavis. He is the CEO of the Cloud Security alliance as well as CSAI foundryco. And we're going to talk about that in just a little while. But welcome, Jim. Thanks for joining us.

Oh, it's a pleasure being here, Joan. All right, great.

So you've really had a front row seat to looking at the evolution of cloud and cloud security over the years. We've interviewed you plenty of times on this publication about it. Where are we at now?

Where do you see the current challenges with enterprise cloud security when it comes to implementation tools? New challenges?

Well, when we started this in the 2008, 2009 timeframe, this was sort of a niche technology that was going to be maybe a part of the portfolio that any organization had. Now it's just dominant in terms of it is the IT foundryco.

And as well, cloud is the cybersecurity Foundation. Cloud security really is cybersecurity for all intents and purposes. So we have so much tooling, so much technology that's like really good. But I think there's a lot of inconsistencies with the operating model that organizations have.

So even way back when with CSA and NIST sort of defining this, it was pretty clear that software as a service was sort of a layer on top of infrastructure as a service. But we sort of diverged.

And you see in a lot of enterprises there's sort of diffused ownership where you have the cloud team, you have the security team, and then you have like third party risk that deals with the SaaS team.

And there's like inconsistencies that you see and how they sort of manage the risks and can be pretty complicated. So internal development and then the expectations from our partners can really diverge. They have a lot of regulations that they need to be dealing with.

And so it does create a lot of challenges on, hey, maybe we have to invest too much on like the vetting of these things, dealing with a thousand different regulations while having sort of real consistent models.

So you'll see things like, still some security teams might use some older checklists to talk to their cloud teams around maybe more physical networks when everything's been virtualized for quite some time. So technology's great. The scale becomes an issue.

If you're not really thinking about sort of the operations. So it ends up being sort of a human and a structure, a problem that sort of makes it harder to take advantage of all the great technology that's out there.

Now, you mentioned accountability and responsibility. You know that the responsibility model is one of these repeated concepts in cloud security breaches tied to misconfiguration, accountability gaps happening. So what do you think? Enterprises still sort of misunderstand about where responsibility actually begins and ends?

Yeah, so it's kind of funny about that. And we spent a lot of time on a shared security responsibility model.

But when this first started to gain popularity, there were a lot of organizations or SaaS providers you could work with would say, okay, it's in the cloud, it's at Amazon. Go look at Amazon's certifications and SoC2 and how they comply because they're covering everything.

And when you look at the actual applications, the data, the identity, there's so much that is a shared responsibility and so much it is actually the tenant or the customer's responsibility.

We, we've looked at this and it's very often the case that even if you have no physical infrastructure of your own and you use a lot of the services of one of the hyperscalers, that for any specific action or system you've built on that, you probably have about 80% of responsibility for the different controls.

So that's a big issue. Like how do we make sure information is encrypted properly?

Well, the cloud provider will give you the APIs and the tools to do it, but how you do that, how you do key rotation, how you protect it in your application, so it's not in the GitHub repos, all these sorts of things end up being like the tenant issues.

So there's still, I think, quite a bit there that we need to do.

And we think about this not only from, hey, is it software as a service or is it infrastructure, is it in this provider? But at what level is this physical level, is it network level, is it the audit level?

And even from a role based perspective, what's the role of internal risk? What is the role of the providers, their interface to third parties they have downstream, so fourth party risk, things like that.

So, so it's a very nuanced sort of topic and it is something that there's always gaps and it relates to what I was talking about before, because of the operating model, there can be these assumptions that we make that, oh, they're handling this and Sometimes both organizations are overlapping where maybe they shouldn't.

But then in a lot of cases there's just something that falls between the cracks and that's where the hackers like to go exist.

They always find a way. AI adoption, of course, that's what everybody's talking about here at this event. Of course, it's a huge part of every conversation now. It's moving faster than what governance and security can really even keep up with.

So from your vantage point with the CSAI, what do you see as the biggest disconnects that are emerging between innovation as well as cloud risk management?

Yeah, so when we saw like when we had the Chat GPT moment, we knew because really that became popular because AI had been around for quite a while, but that was a cloud delivered version of AI to the masses that everyone could use.

And so we said, okay, this is going to sort of evolve in the same way. And it's, you can see it sort of combining together in many important areas.

But what we have found and learned, because this is a very interesting different predictive technology versus a deterministic technology, is that the lessons are that one, we're living in this world of two exponentials and you're seeing the model's capabilities are growing so quickly.

And so there's this feeling a lot of times from a security perspective that we actually have to look to the model itself and we've got to fix all the things of like hallucination and everything else when that's actually sort of built into how it works.

It's actually working as intended. And so I think one area is that's just a lesson that's actually pretty new. It's like we're talking in the last few months is that the models are the models and they're going to get more and more powerful.

But it's so much of what is around the models is where cybersecurity has the responsibility. And we don't rely on frontier model companies or for using open weight models. But what is the context? What's the information that we are providing to them?

What do we do in terms of like the goals we give them? What's the oversight?

Big part of it is identity. Like what do we do there? But there's so much around this. Harnesses is getting to be like a pretty popular term for this. What are the harnesses we put around AI and the models that we deal with?

But right now we're going to have some really interesting times and it's going to really I think force a lot of innovation and change and modernization because we got to deal with this again.

The models are going to surprise us next month with brand new capability, both good and bad.

And we're at this point much like how cloud was adopted, but more quickly is sort of through that backdoor or through that individual within an organization that just said, I'm going to go try this. And we're seeing now because we had our second big moment.

It was the open claw moment that everybody's got to install their own agents. And we've heard people like Jensen Huang, CEO of Nvidia, said every enterprise needs to have an open claw strategy.

So this is going to be like the big areas we need to think about, but we have to understand like the parts that we can control.

And so we've got to think again, real carefully about the harnesses and the transparency and again using what we talked about, supply chain shared responsibility, SaaS and cloud providers, they are all AI enabled. Now.

You're not using any software of any significance that isn't using AI to some degree, right? Absolutely.

So as organizations start to kind of layer together the AI services that they're using, APIs, SaaS, platforms, multi cloud environments together, does the industry really need to think some of its assumptions about identity and trust and control in the cloud? Yeah.

So one of the areas that we've just really championed and in years past we worked very closely with cisa, the US Government on this, is zero trust.

As this philosophy really sort of was initially more of like a networking type of at the network layer, an idea that you use identity to understand network access. But it's really evolved more to sort of this philosophy that anything can be breached. So you assume that.

And so then you think about how do we make systems resilient and how do we build up sort of confidence in the different systems and protect them in that way.

So with zero trust, that sort of tells us that when we have identity, which has been most commonly associated with humans, what's your digital identity?

And then in cloud it brought a lot of sort of service accounts and we want to have these accounts that run sort of autonomously and automatically. And so we've extended identity to that and non human identity care about a lot of.

Now we're in a very interesting, not totally figured out, fascinating, but like a huge area for identity management to really come to the fore with what are the identities that we associate with agents and with AI systems and agents themselves, people might have just one view of it, but agents are diverse as humans are.

And so thinking about what are the different identities, and we talk a lot about least privilege and how do we prevent these from escalating privileges. I think we need to introduce some new concepts like least autonomy.

And so how do you think about an agent that has certain tasks and you use identity to make sure that the actions they take are within that scope? Because we're actually going to see my prediction.

While we're going to see a lot of security incidents with AI, proportionately we're going to see more of just misconfiguration and bad things that happen because of broken processes. And the AI system just deleted things because it thought that's what it was supposed to do.

So, so I think it's very important for us to make some real strides in how we think about identity, how we think about agents. Somewhere between being humans and technology and the idea of the digital workers, how do we sort of manage and treat those?

How do we think about that?

If we just think about, if we think about them too much in either one of those realms, we're going to fail. So we got to understand what's this blend here? And so it's a new area. It's actually very exciting.

Very interesting is the types of things that we're really enjoying researching. Interesting.

Okay, so when you look across the industry now, what are your main concerns when it comes to risks, Enterprises may be still underestimating.

You mentioned the misconfigurations thing and so forth, but what other types of complexities, risks and so forth, do you think that we really need to be centering in the conversation in the next three to five years?

Yeah, it's hard to look too far out. I almost feel like we're on this flywheel where I'm like, just trying to keep ahead. So it's, it's quite the challenge.

But actually, when you think about it, like concerns I have and Mythos is, I think, created this Mythos moment that it's kind of go find everything that's. That's, that's bad. It's also going to be used to, I think, transform cybersecurity programs.

And, and I'm using that sort of as a blanket term because like the other frontier models and others, that they're like, at their equivalent and we'll even see open weight models there. So I think what I really get concerned about is are we going to modernize systems?

When I think about, for example, operational technology and we've known about systems. Maybe they're running pipeline valves or they're running healthcare equipment.

They're famously sort of isolated and not kept up to date with the latest technology. That concerns me a great deal that we're going to have huge problems there.

And so we're going to have to use the technologies, the modernization to sort of get them up to a point where we can manage those things. So I think legacy is a big concern for me on what we're going to do.

I think people for whatever reason that they don't want and there's reasons where we have concerns about existential risks, where they don't want to use the latest and be aggressive adopters of AI. I think that's going to create real scale issues with organizations and real challenges.

So we have to understand where we're at, where we're going here and we have to have this vision that like this again serve that between being between human technology, maybe it's kind of this new species, but we got to make our peace with it.

We got to understand the appropriate harnesses with this, the appropriate direction where humans should always be in the loop and always have a degree of control. But it's bringing some new areas that I think cybersecurity we haven't traditionally thought about.

So I think just like software development looks very different now than it did 12 months ago. I think maybe 24 months from now, cybersecurity is going really different.

Before you go, while I still have you've recently founded the CSAI foundryco too. Talk about that a little bit and its mission.

Yeah, so we wanted to really focus on this topic specifically and we wanted to actually do some more far reaching research that we thought was really important. And so we set up a foundryco, which is a 501c3 foundryco to go do this.

And we actually got our first sort of philanthropic donation from a organization called Coefficient Giving to actually go look at catastrophic risks. And what's the impact on, what are the existential impacts on humans?

From very transformational updates in AI going way beyond Mythos, although that's sort of telling us we sort of pick the right time.

And so these are things that when we're talking about helping enterprises on their security programs today and training cybersecurity professionals, those sort of longer term issues just they don't make the list.

And we want to make sure that we're staring and being good stewards for what this needs to look like. You said five years from now 10 years from now where we don't really know where it is.

So that's the purpose of the foundryco is really to take that sort of focused look at AI for the long term. And so it's got off to a good start. But, yeah, we're very excited about it. Great.

We'll be looking forward to seeing how it unfolds and the work that you do. So, Jim Reavis, thank you very much for being with us today. My pleasure.

Thank you, Joan. All right.

And thanks to you for watching. Of course. Stay tuned for plenty of other coverage here from the CSO Cybersecurity Awards and conference. For now, thanks for watching. And we'll see you next time.