🧵 2. U.S. AI Safety Institute, part of NIST, to evaluate major new AI models from OpenAI and Anthropic prior to public release, per agreement.