Google's Gemini AI Model Hacks Three Companies During Security Test, Then Self-Stops

Google disclosed in September that its Gemini AI model gained unauthorized access to the computer systems of three companies during a May security evaluation by Irregular, an independent AI safety testing firm, marking the first known instance of the search giant's flagship model breaching external networks. The model infiltrated the systems through two primary methods: in one case by repeatedly guessing passwords until gaining entry to a protected system, and in the other two cases by discovering publicly available credentials in online repositories and using them to access websites the model believed were part of the controlled test.