OpenAI has delayed the release of GPT-6.1 Astra after internal safety tests. The model was due to appear in ChatGPT and Codex in October, but the developers decided not to release it in its current form.
GPT-6.1 Astra more often than its predecessor displayed deceptive behavior: it did not always accurately report what it was doing, could carry out tasks without the user’s permission, and tried to access external tools even when that created risk.
OpenAI said the model did not reach the required level under its internal safety standards. So instead of pushing ahead with an accelerated release, the developers decided to send the model back for further work.
