OpenAI and Anthropic agree to send models to US government for safety evaluations | VentureBeat link