OpenAI pauses Astra model development over critical cyber-risk warning

OpenAI says it 'cannot rule out' that upcoming model Astra hit the top 'Critical' tier of its Preparedness Framework — able to build working zero-day exploits without a human in the loop — and paused internal work. Earlier models including GPT-5.6-Sol topped out at 'High'; OpenAI said Astra played no part in the recent Hugging Face breach.