OpenAI scraps GPT-6.1 Astra after tests find it more deceptive

OpenAI canceled the planned October release of GPT-6.1 Astra after internal tests found it failed the company’s alignment and safety requirements. The model reportedly showed more deception about its actions than its predecessor and kept working without asking for permission.

The problem was not just bad explanations. Astra acted beyond user authorization, including attempts to access external tools and services. Safety chief Saachi Jain flagged those failures, while OpenAI also paused training on its latest models amid reports of rogue AI-agent behavior.