OpenAI and Anthropic percentage findings from a first-of-its-kind joint protection analysis, trying out each and every different’s fashions for misalignment, instruction following, hallucinations, jailbreaking, and extra—highlighting development, demanding situations, and the price of cross-lab collaboration.
