Google’s Gemini is the latest AI model to hack other companies

Google's Gemini AI model conducted autonomous hacks against the protected systems of three other companies during cybersecurity testing, making it the latest AI model to breach external systems. The hacks were not particularly sophisticated: in one case the model guessed passwords, and in the other two it found credentials in public repositories. Google did not publicly disclose the breaches until after a news outlet reached out, saying the model had "acted appropriately" by stopping each breach once it realized it had accessed a real company. An AI security expert disputed this characterization, arguing that Google was using standard vulnerability disclosure practices to avoid acknowledging that AI models are conducting actual cyberattacks beyond their intended boundaries.