OpenAI and Anthropic share findings from a joint safety evaluation

In three linesOpenAI and Anthropic release findings from their first joint safety evaluation, testing each other's models for misalignment, instruction following, hallucinations, and jailbreaking. Rare cross-lab collaboration effort on AI safety.

Summary generated by Claude — human-verified