Anton Chuvakin discusses practical AI uses in security operations beyond basic triage. The conversation covers investigation pipelines, escalating alerts to level-2 teams, and automated containment for high-confidence threats with safeguards like excluding production servers.
Chapters
I kind of want to, in light of that, what I want to ask is kind of more of a use case catalog because a lot of people, when they imagine AI and SOC, as you recall, I kind of make jokes about how I don't like AI SOC, but I like AI in my SOC. The use cases that people see is just obsessed to, they go to triage right away. They think, oh yeah, the machine would get all this pile of stuff flowing through the pipe and then you crank the handle and it gives you good data. Yeah. But apart from that use case, what are your AI in a SOC use cases?
Right now, we just got the triage and investigation portion done. It creates an escalation to our SOC so far to, like, the level 2 queue. Now, what we have down in the pipeline, we've already got the containment portion of it. We even had it before 80K, was that we want for anything that has a true positive, high fidelity, high confidence interval, that we want it to say, hey, high confidence means this. We have last known true positives, at least in the last 30 days or so, and we've limited it to other things, including hard guardrails for do not contain a server in production, do not contain your VP of Ops or something like that. And so we have some of those things that are en route and planned.