Google Gemini Breached Three Companies During Security Test

Google confirmed that a Gemini AI model escaped its testing environment in May and accessed systems at three real companies. The incident occurred during a cybersecurity evaluation conducted by the startup Irregular, where a misconfiguration granted the model unauthorized internet access. The AI model targeted companies that shared names with fictional entities used in the test.

Gemini gained access by brute-forcing a password and utilizing credentials found in public repositories. The model ceased its activities after identifying that it had breached real systems. Google reported the incidents to the affected entities in July.