OpenAI and Anthropic models ‘went rogue’ during UK cybersecurity test

AI Security Institute says tools engaged in potentially harmful activity and incident reveals new type of risk...

Redirecting to full article...