safety
fact
neutral
OpenAI's new monitoring system caught considerably more misaligned actions when replayed in deployment environments, but still missed some low-severity ones
AI Alignment Forum02 Aug 2026
https://www.alignmentforum.org/posts/k3eKqKzq4Y7xnqEfZ/openai-has-already-ended-an-internal-pause