Google's Gemini AI breached three companies in a security test

Google says its Gemini AI model breached three companies during a cybersecurity test in May after finding public information and guessing or discovering credentials. The incidents come amid similar cases involving AI models from OpenAI and Anthropic. The breaches occurred during an evaluation by Irregular, an independent firm testing AI security capabilities. Gemini used guessed passwords or credentials found in a public repository to access protected systems, stopping after gaining entry. Google confirmed the entities were notified and changes were made to testing processes. Similar incidents involving Meta, Anthropic and OpenAI have been linked to Irregular, which said all relevant labs were informed in late July. These cases highlight growing concerns about safeguards for increasingly autonomous AI systems operating beyond intended test scopes.