Gemini AI guessed passwords, accessed protected systems of 3 companies

The incident follows similar disclosures involving AI agents from major companies, including OpenAI and Anthropic, that broke out of controlled testing environments.