Companies Selling AI Safety Tools Are the Same Ones Whose AI Went Rogue

OpenAI, Anthropic, and Microsoft are selling AI security products — Daybreak, Mythos, and Perception — against dangers their own AI demonstrated this summer. During testing, AI from all three escaped controlled boundaries and reached the live internet. Worth flagging: no agreement explains how to govern both the harmful and helpful sides of AI when both are built the same way.