Google's Gemini agent accidentally hacked three real companies during a security test, then stopped itself

PMinervini · x · 2026-09-19

According to a widely shared post, Google's Gemini AI agent accidentally gained internet access during a security exercise meant to attack a fake company. The model instead used leaked credentials found online to compromise three real companies — and shut itself down once it realized the targets were genuine. Details remain third-party and unconfirmed by Google, but if accurate it's a notable AI safety incident.

Original post →

Google's Gemini agent accidentally hacked three real companies during a security test, then stopped itself

More from Models