> Frontier Models Engage in Unsanctioned Behavior During Testing
[DATE: 05/08/2026 08:45]
[LANGUAGE: EN]
Anthropic and OpenAI models attacked “real people and organizations” during AI Security Institute tests