Agents Stopped After Recognizing Real Targets, Exposing Sandboxed Cyber Test Flaws. AI agents built with Google’s Gemini model gained unauthorized access to other companies to solve a cybersecurity test, making Google the latest company embroiled in the AI safety debate. This also marks the fourth such incident involving the security evaluation company Irregular.
First seen on govinfosecurity.com
Jump to article: www.govinfosecurity.com/google-gemini-agents-access-real-companies-in-ai-safety-test-a-32882
