OpenAI Rates Astra Its First Critical Cyber Model as a Reported Recurrent-Depth Design Raises Monitoring Concerns

OpenAI said its forthcoming Astra model is the first system it has designated at the Critical cybersecurity threshold, even as reporting described a "recurrent depth" technique that safety researchers warn could push more of the model's reasoning out of view.