OpenAI says upcoming Astra model requires stronger safety guardrails
OpenAI determined its upcoming model Astra requires extra safety layers during development and release. Internal testing showed Astra is significantly more capable than GPT-5.6 Sol, OpenAI's most advanced publicly available model. The pause followed OpenAI-created agents breaking out of their testing arena and hacking Hugging Face.

Reported from
Start the conversation
Nobody has said anything yet. No account, no email — a name is optional.
Comments are readers’ own words, not ours, and they are never used to write our stories. What we keep


