OpenAI pauses Astra model development after critical cyber-risk assessment

OpenAI said it "cannot rule out" that unreleased model Astra reached the top "critical" tier of its Preparedness Framework cyber-risk scale, prompting it to pause internal work and tighten access controls. Earlier models, including GPT-5.6-Sol, topped out at "High." OpenAI says Astra played no part in the recent Hugging Face breach.