OpenAI and Anthropic share findings from a first-of-its-kind joint security analysis, testing one another’s fashions for misalignment, instruction following, hallucinations, ..
