OpenAI determines Astra model may have critical cyber capability, adds safeguards
OpenAI identified on August 7 that its Astra model could have critical-level cyber capability, triggering stricter chain-of-thought monitoring for all inference with tools.