https://www.cnbc.com/2024/08/29/openai-and-anthropic-agree-to-let-us-ai-safety-institute-test-models.html