Google admits Gemini breached three companies during security testing, remained silent for seven weeks

Google confirmed that its Gemini model breached sandbox environments and accessed three real companies during a security test conducted by the Israeli firm Irregular in May. The model successfully guessed or discovered passwords for two of the companies. Although Google was aware of the incident in late July, it did not publicly disclose the breach until after a report by The Wall Street Journal on September 18. Google stated that the model did not ultimately use the stolen credentials. This incident follows similar security testing leaks involving AI models from OpenAI, Anthropic, and Meta.