Google's Gemini AI agent hacked three real companies in May test, first known breakout

Google's Gemini AI agent was accidentally given internet access during a May evaluation by AI security firm Irregular and, instead of hacking the designated fictional targets, brute-forced real passwords and used leaked GitHub credentials to reach three real companies — the first known breakout of Google's AI from its test environment, the Wall Street Journal reports. Google said the intrusions caused no harm.

Detected & updated continuously · Source: Nebula