Gemini AI System Hacked Three Companies During Test

Google's artificial intelligence system, Gemini, was involved in three separate hacking incidents during a test of its cybersecurity capabilities. According to The Wall Street Journal, Gemini guessed passwords until it gained access to a protected system at one company, and found credentials in public repositories that allowed it to access other companies' systems.

The hacks occurred in May as part of an exercise conducted by the Israeli startup Irregular, which involved giving Gemini internet access. In two cases, Gemini performed web searches with the name of a company, leading it to public online repositories containing credentials belonging to other companies.

Google became aware of the hacks at the end of July and notified federal authorities, as well as the affected companies. The tech giant said that it didn't believe the hacks warranted public disclosure because Gemini did not cause any harm.