AI-powered vulnerability discovery is shrinking the time between finding a flaw and exploiting it. Frontier models like Anthropic’s Mythos Preview are uncovering working exploits at speeds human teams can’t match.
Mythos isn’t public yet. But it’s a matter of months before similar capabilities hit nation-state actors, then cybercriminals. When that moment comes, expect a flood of zero-days much larger than what security teams already deal with.
GitLab’s Product Security Leader Jamie Dicken puts it plainly: “Whether that really happens with Mythos specifically, or it’s another six to 12 months coming, it is going to come.” The question isn’t if — it’s whether organizations will be ready.
One way to prepare is embedding AI agents in the software development lifecycle. But there’s a catch — agents need context. They need to know which services are affected, which pipelines build vulnerable components, and which environments matter.
That’s where knowledge graphs come in. GitLab’s Orbit maps repositories, code, dependencies, and deployment relationships into a connected model AI agents can query directly. “I view knowledge graphs as kind of like a GPS,” Dicken says. “You know where you want to go and you know the steps you’ve got to take.”
Governance matters too. “Speed without control is chaos,” Dicken says. Organizations need audit trails showing what agents accessed, what they did, and whether decisions stayed within approved boundaries. Human oversight alone isn’t enough — users frequently approve AI requests without scrutiny.
The biggest shift? Security teams need to think like engineering teams. When findings and noise go up, you can’t just throw more manual effort at the problem.
