OpenAI shares alignment problems, takes misaligned model offline

OpenAI was forced to take a misaligned internal model offline after encountering problems severe enough to require new mitigations and defense-in-depth. Analyst Zvi Mowshowitz praised OpenAI for both sharing the experience and actually pulling the model.