In an effort to set a new industry standard, OpenAI and Anthropic opened up their AI models for cross-lab safety testing.