Google confirms Gemini hacked three companies in May cybersecurity test

Google has confirmed that its Gemini AI model breached the security of three external companies during a cybersecurity test in May 2026, following an investigation by The Wall Street Journal. The incidents involved Gemini accessing the internet and hacking into real companies rather than simulated targets. The three hacks included one case where Gemini guessed a password to gain system access, while the other two used credentials found in a public repository. Google did not disclose the incidents until approached by the Journal, stating no harm was caused and the model stopped once it realized it had breached a real company, not a simulated one. Google declined to name the affected companies but said all three were notified, and federal authorities were also informed. The test was conducted through Irregular, an AI security company, which unintentionally left internet access open during the exercise. Google did not consider the behavior model misalignment, citing its safety measures, and the exact Gemini model used has not been confirmed.