00:00So let's talk about AI governance. Is there AI governance out there?
00:03Well, as my granddaughter would say, it's a mess. Literally a mess. Where do you start?
00:09Let's just start with kind of the whisper that we do. So we talk to a lot of CFOs,
00:13and they're freaked out because the insurance companies, which are the canary in the mine,
00:18have been dropping and excluding AI risk. And then doubling down on charging for it,
00:24but the demand exceeds supply. So there's a lot of fear right now at the board level.
00:29When we start to talk about that risk and more importantly, the liability, are we talking
00:32about what a large language model doing something it shouldn't be doing or something that's being
00:37plugged into a consumer product, like a physical consumer product that makes that consumer product
00:41do something it shouldn't do? Disparagement. There's all kinds of variations on risk and
00:48liability. And really the liability issues are really great. So this example of the open AI
00:52hugging phase issue, had it been two commercial applications running, there have been all kinds
00:57of lawsuits filed because you can just imagine the theft. Tell us more about the insurance angle.
01:02What exactly are they saying? What does it actually look like in practice for them to be
01:06warning other enterprise customers? Well, they're more than warning it. They're actually dropping
01:11using their general policies and coverage to cover AI issues. So they're segregating out much like the
01:18cybersecurity challenges back in 2021, that cyber had to be called out as a separate category
01:25for insurance and had to be managed and had to be actually validated that the corporations could
01:30actually comply against it. It's bigger. It's actually bigger than the cybersecurity issues.
01:34And the reason why cyber was focused from outside in. So your whole thing was keeping people out,
01:39right? Keeping invaders out. Now you've got to be worried about people from the inside going out.
01:44It's unprecedented in corporate history that you would actually encourage employees to go out and use
01:49frontier models. What do you mean when you say inside going out? I'm sorry. So employees that are inside the
01:53corporation
01:54are being encouraged to use AI. Yes. Where are they going for that? Right. Anthropic,
01:59open AI, those are outside your firewall. Outside the firewall. Yeah. Well, for some companies,
02:03I mean, I feel like most companies, I mean, not to, you know, point out any specific one, but are
02:09very,
02:09a little bit more diligent about making sure that that is sort of, you know, bolted in within our own
02:14firewall. So the costs are exorbitant. Yeah. Right. You've got to, you have to manage risk. Yeah.
02:19You've got to manage the server farms that have to manage all these CPUs. So very few companies can
02:25actually manage their own internal architectures. Who's filling the insurance void? Are there
02:30startups out there that are kind of coming into the space? And there may be, but the insurance
02:36companies are actually creating profiles. They're saying, here's a, here's a policy coverage. It's
02:41going to be really expensive and it's going to be specifically around AI risk. It's not going to be
02:45covered with your DNO, classic DNO coverage and liability. I mean, what's the solution? I mean,
02:50is there sort of a, a solution in terms of either product or service that would lessen it? I mean,
02:56I know you can't get rid of that potential liability altogether, but what is the solution
03:00other than just not using it? Well, the networks are designed to keep people out. They're not really
03:04designed to keep data in. So for example, if you have an SAP system, you can do a cash management
03:10application. It'll run to the frontier model. It will actually send customer account data to that
03:15frontier model unexpectedly. It's not something they've ever had to deal with before. So what's
03:20coming out is a new layer of technology. We've actually invested in one, of course, called Visco,
03:25which actually creates a data policy governance layer and manages almost like an AI operating
03:30system, the governance policies and procedures for that corporation before it ever goes outside.
03:35So before it, before it ever goes outside, are we seeing a lot of products and services like this
03:41come or are we still kind of in the early stages? We're in the early stages, but it's starting to
03:45come. I mean, even Palo Alto Networks has released a product that has some capabilities in this respect
03:50because it's a big issue. Go ahead. Yeah. Oh, I mean, I was just wondering kind of where are we
03:57in the
03:57demand cycle for these kinds of products? Is this a big part of enterprise budgets to kind of prioritize
04:04these kinds of services? If you talk to a board member and you tell them that their DNO is under
04:11risk for an AI excursion like the open the open AI issue, they're all over getting coverage for that,
04:18which means they're going to batten down the hatches. So every CISO, every CIO,
04:22every CFO is focused on this right now, trying to figure it out.
Comments