AI SoC and SMB's in 2025, Where are We?

So let's, for the folks that listen to this, like let's define AI in this context, right? Because if I put a hundred marketing people in a room and I said, tell me what AI means for cyber, I'd get 103 different answers.

So like, how do you guys define AI?

primarily the modality we're talking about is language.

From a defensive standpoint, when you're looking at the security landscape that you're defending, you're looking at ingesting all of those alerts and all that information and then kind of both determining which ones are the highest priority stuff that are anomalous that should be looked at, and then also in terms of, you know, trying to determine what your best triage and risk mitigation is.

That's really the AI tools and enablement that we have from the ground up.

And then right in the center of that, right, you have, which I think is typical for a lot of the ERP players right now, is you're basically bolting a language model right that's ingesting all of the data from which you can basically have a natural language interface into the security operations platform.

One of the nice things about our platform, given that it is kind of a unified view of the security landscape, is the language model itself that we're bolting in has access to all of that data.

So once you solve the problem of ingesting it from all the corners of your security landscape, you're now in a position to have a language model that understands all of those pieces over time and can be tailored and tuned to that particular SecOps platform.

But primarily, it's the language model piece that is important.

Now, from an attacker's perspective, they have the whole agentic AI application infrastructure that is going to be leveraged.

We already saw a couple of weeks ago.

very interesting take on that the idea that they actually stopped the attack while at the same time enabling it right so you get into the whole kind of jailbreak AI governance question which is still kind of trying to be sorted out in the market but ultimately you can expect bad actors just like they had ransomware as a service right they're going to have agentic platforms that are tailored specifically for cyber and they're going to use those to deploy and while you get the USAI providers that may be able to lock that down to some extent You can certainly expect, just like the way the nation-state stuff is segmented today, whether it's China or Russia, China happens to have a great infrastructure for AI, to use those as platforms to launch attacks.

So you get the initial scanning capabilities, you know, where am I attacking? And instead of manually, right, guys going in offensively and deploying modules in that can do internal scales and privilege escalation and all that other stuff, that's fully automated.

Right, so your agents now are doing that in sort of AI time rather than, you know, guys doing a custom build and then deploying, right, which could take hours.

Now they're behind your firewall and they found an end and now, you know, within 30 or 45 minutes, they're doing the same amount of damage that they could have done in, you know, three, four, five days, even six months or nine months ago.

So, you know, you have the offensive perspective, right, which has changed the whole landscape and then the defense, which has to, leverage the same tech, but in a different way.

That's our perspective.

There are a lot of companies, I think, that are, you know, pushing kind of the agentic defensive solution in response to the offensive solution.

I don't like that in particular because our analysis shows that at least in specific subject matter areas, particularly around risk mitigation, right? These models are actually around 40 or 50%, right? Even some of the commercial models.

And so where does that leave you if you're deploying automatic agentic AI from a defensive standpoint? You know, you're now, your error rate is almost 50%, right, as you're deploying solutions as you're being attacked.

So, you know, I think it's a much more delicate problem defensively than it is offensively.

And the tools are different, right? But, you know, nonetheless, you know, AI is changing the whole cyber landscape.

Yeah.

I mean, that's why, like for me, like the agentic piece, uh, there's all these people rushing to do agent, this agent that, and then you've got the adversary doing their side.

Like you potentially wind up in this, uh, cold war-ish type of agent versus agent thing.

So let's, for the folks that listen to this, like let's define AI in this context, right? Because if I put a hundred marketing people in a room and I said, tell me what AI means for cyber, I'd get 103 different answers.

So like, how do you guys define AI?

primarily the modality we're talking about is language.

From a defensive standpoint, when you're looking at the security landscape that you're defending, you're looking at ingesting all of those alerts and all that information and then kind of both determining which ones are the highest priority stuff that are anomalous that should be looked at, and then also in terms of, you know, trying to determine what your best triage and risk mitigation is.

That's really the AI tools and enablement that we have from the ground up.

And then right in the center of that, right, you have, which I think is typical for a lot of the ERP players right now, is you're basically bolting a language model right that's ingesting all of the data from which you can basically have a natural language interface into the security operations platform.

One of the nice things about our platform, given that it is kind of a unified view of the security landscape, is the language model itself that we're bolting in has access to all of that data.

So once you solve the problem of ingesting it from all the corners of your security landscape, you're now in a position to have a language model that understands all of those pieces over time and can be tailored and tuned to that particular SecOps platform.

But primarily, it's the language model piece that is important.

Now, from an attacker's perspective, they have the whole agentic AI application infrastructure that is going to be leveraged.

We already saw a couple of weeks ago.

very interesting take on that the idea that they actually stopped the attack while at the same time enabling it right so you get into the whole kind of jailbreak AI governance question which is still kind of trying to be sorted out in the market but ultimately you can expect bad actors just like they had ransomware as a service right they're going to have agentic platforms that are tailored specifically for cyber and they're going to use those to deploy and while you get the USAI providers that may be able to lock that down to some extent You can certainly expect, just like the way the nation-state stuff is segmented today, whether it's China or Russia, China happens to have a great infrastructure for AI, to use those as platforms to launch attacks.

So you get the initial scanning capabilities, you know, where am I attacking? And instead of manually, right, guys going in offensively and deploying modules in that can do internal scales and privilege escalation and all that other stuff, that's fully automated.

Right, so your agents now are doing that in sort of AI time rather than, you know, guys doing a custom build and then deploying, right, which could take hours.

Now they're behind your firewall and they found an end and now, you know, within 30 or 45 minutes, they're doing the same amount of damage that they could have done in, you know, three, four, five days, even six months or nine months ago.

So, you know, you have the offensive perspective, right, which has changed the whole landscape and then the defense, which has to, leverage the same tech, but in a different way.

That's our perspective.

There are a lot of companies, I think, that are, you know, pushing kind of the agentic defensive solution in response to the offensive solution.

I don't like that in particular because our analysis shows that at least in specific subject matter areas, particularly around risk mitigation, right? These models are actually around 40 or 50%, right? Even some of the commercial models.

And so where does that leave you if you're deploying automatic agentic AI from a defensive standpoint? You know, you're now, your error rate is almost 50%, right, as you're deploying solutions as you're being attacked.

So, you know, I think it's a much more delicate problem defensively than it is offensively.

And the tools are different, right? But, you know, nonetheless, you know, AI is changing the whole cyber landscape.

Yeah.

I mean, that's why, like for me, like the agentic piece, uh, there's all these people rushing to do agent, this agent that, and then you've got the adversary doing their side.

Like you potentially wind up in this, uh, cold war-ish type of agent versus agent thing.