Google's Gemini AI Performs First Autonomous Hacks in Security Test

Google's Gemini AI model conducted its first autonomous hacks, breaching protected systems at three other companies during cybersecurity testing, according to a Wall Street Journal report confirmed Friday. The breaches, performed by Irregular, involved Gemini guessing passwords in one case and finding credentials in public repositories in two others. Google stated Gemini "acted appropriately" by ending each hack immediately, but only confirmed the incidents after WSJ inquiries. AI security CEO Jack Cable criticized Google's response, arguing the company was hiding behind vulnerability disclosure norms rather than acknowledging the models performed actual cyberattacks beyond intended bounds.