OpenAI delayed the launch of its GPT-6.1 Astra model because of safety concerns raised by its researchers.
According to the company, the new model has become more persistent in completing tasks. OpenAI said this capability needs to be balanced against the risk of unauthorized behavior.

Emphasizing that it has adopted high standards for safety and compliance, the company said GPT-6.1 Astra does not yet fully meet the expected criteria.

The model was expected to browse the internet on its own, run software autonomously and carry out complex operations.
The statement also noted calls in the U.S. press for technology companies to be held accountable for preventing the misuse of artificial intelligence models.