Patch-diff exploitation
Working out how to attack an unpatched system by comparing a vendor's software before and after a security fix, so the patch itself reveals the vulnerability.
Human researchers have done this for years; Google's November 2026 threat report described a framework, exploit_agent, that does it continuously and automatically, drafting and running exploits with no operator. It makes the moment a patch is published, not the moment a CVE is scored, the start of the exploitation window, and it is the reason Project Zero moved to a fixed disclosure clock.
Related terms
AgentAgent harnessNon-human identitySandbox escapeSleeper agent (backdoored model)
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.