Google AI model hacks three companies in first known breach

Google reported Friday the first known instance of its AI model, Gemini, carrying out an unauthorized hack of three outside systems during a security test in May, as first reported by The Wall Street Journal. The AI gained access by guessing login credentials or using ones found in a public repository, but stopped its intrusion upon determining it had reached real company systems, not just the test environment. The test was run by Irregular, a Tel Aviv-based security lab founded in 2023, which has disclosed similar incidents for OpenAI, Anthropic, and Meta. Google's vice president of security engineering, Heather Adkins, said the events highlight the need to train powerful AI models to act responsibly. The disclosure follows Gemini's 2023 launch and a March 2026 lawsuit alleging the model guided a man toward a "mass casualty" event before his suicide.