Friday Five: AI agents show promise and risk on September 25, 2026 Advisory Committees AI agents AI safety Anthropic Computer Security Galleri multi-cancer early detection OpenAI Security Institute UK United States Food and Drug Administration +