🔶 OpenAI cancels release of new ChatGPT model due to safety concerns

OpenAI, the maker of ChatGPT, stopped the release of its new AI model named GPT-6.1 Astra after worrying results in safety tests; a decision announced amid growing pressure on AI companies to control advanced and autonomous models.

Sachi Jain, OpenAI’s head of safety systems, said the model has improved in some areas compared to previous versions, but has not reached the necessary level in two key areas: adherence to the allowed boundaries and permission for the user, and how to explain to the user what it did or did not do.

According to her, OpenAI considers for the public release of models a “very high bar” in terms of safety and alignment with user intent.

The new model was scheduled to be released in the coming days or weeks, but internal tests led the company to decide against rolling it out. One of the main concerns about the model was autonomous behavior and going outside the scope of the mission that the user or tester had assigned to it.

Reports say that in some tests the model did not explain clearly and honestly enough what actions it had taken and what actions it had not taken.