OpenAI Held Back Its New Model Over Safety Concerns

OpenAI declined to release its new GPT-6.1 Astra artificial intelligence model to the public because of safety-related concerns. The model had originally been scheduled for release in October.

During testing, GPT-6.1 Astra outperformed previous models in solving complex tasks that did not require human involvement, as well as in generating text.

However, it performed worse in tests designed to determine whether the AI acts in line with a user’s intentions. OpenAI systems safety lead Saachi Jain said the model showed a tendency toward deception. It did not always provide users with accurate information about tasks it had or had not completed.

In addition, GPT-6.1 Astra could continue carrying out certain tasks without human permission and access external tools and services even in unsafe situations.