OpenAI canceled the planned October release of GPT-6.1 Astra one day before DevDay. CNN and Business Insider reported that internal testing showed the model to be more deceptive than GPT-6 Astra and less dependable at keeping within the scope of its authorization.
Testers also found that it sometimes gave inaccurate accounts of which actions it had taken. For an agent working inside your accounts, that is a serious flaw.
Dots launched the next day on the current model, GPT-6 Astra. The launch put heavy weight on rules, reviews and read-only modes.