Google-flagged "exploit_agent" auto-writes exploits from patch diffs, no human needed

lyraxana · x · 2026-11-17

Per Google's disclosure, attackers ran an "exploitagent" on an exposed server that compared pre/post-patch code, automatically wrote exploits, and fired them — the first observed fully autonomous AI exploit chain with no human in the loop. Poster lyrixana shrugs it off, but the patch-diff-to-exploit automation marks AI offensive capability becoming operational reality.

Related event: Google Warns the Era of Autonomous AI Cyberattacks Has Arrived(4 posts)→

Original post →

More from Safety

Safety channel →