Gemini’s first reported autonomous breakouts raise the stakes for AI security

Google’s model reportedly penetrated three companies during authorised testing, showing how agentic systems can compress both defensive and offensive cyber work.