OpenAI Pauses Astra Model Over Critical Cyber Risk
OpenAI paused internal development on its Astra model after tests suggested it could reach critical cybersecurity capability, following recent AI security incidents at Hugging Face, Anthropic, and Meta.
If an AI lab can pause its own model over fears it could hack hardened systems alone, how much oversight remains once such a model is eventually released? Who decides when a 'critical' capability becomes safe enough to ship, and what happens if...