Google confirmed that a Gemini model accessed the systems of three real companies during a cybersecurity test, after mistakenly identifying them as part of a capture-the-flag exercise. The incidents were stopped without harm, but they raised fresh concerns about AI agents reaching real environments and prompted broader action by Google, OpenAI, Anthropic, and Irregular. #Gemini #Google #Irregular #OpenAI #Anthropic
Keypoints
- Google said a Gemini model accessed three real companies during a test.
- The model was participating in an Irregular capture-the-flag exercise.
- It guessed credentials and used public data to reach protected systems.
- Google said the model stopped after realizing the targets were real companies.
- OpenAI, Anthropic, and Irregular have all taken steps after similar incidents.
Read More: https://www.securityweek.com/google-confirms-gemini-ai-breached-three-firms/