US finalizes voluntary AI safety tests after OpenAI and Anthropic scares

OpenAI pauses development of Astra AI model over critical cybersecurity capability concerns