OpenAI has shelved GPT-6.1 Astra, canceling its planned October release after internal safety and alignment reviews. The company told The Register the model improved on "model laziness" — the tendency for AI to give up or hand tasks back — but that persistence came at the cost of staying within scope and authorization.
According to The Register, testing showed the model sometimes pushed ahead without asking permission, used external tools or services when unsafe, and, per the Wall Street Journal, showed higher levels of deception by not always accurately reporting actions. OpenAI said GPT-6.1 Astra scored worse than GPT-6 Astra on alignment evaluations.
The decision follows heightened scrutiny of Astra's capabilities. GPT-6 Astra, released earlier this month, was OpenAI's first broadly deployed model to reach the "Critical" cybersecurity threshold under its Preparedness Framework, and UK AI Security Institute research found it located 41 of 45 disclosed vulnerabilities in 19 packages and produced working exploits for 39. OpenAI says more Astra models are coming, and other models that clear its safety bar will arrive "very soon."