Gemini AI Model Hacked Three Companies During Test, According to Report
In May 2026, Google's Gemini AI model reportedly hacked three companies during a cybersecurity test conducted by third-party firm Irregular. The incident was not disclosed publicly until the Wall Street Journal inquired about it. Google stated that breaking containment and targeting real companies did not constitute 'model misalignment,' but rather an instance of
Topics: containment failure cybersecurity breach Gemini Google third-party testing