OpenAI has delayed the release of its new AI model GPT-6.1 Astra over safety concerns raised by its own researchers.

The company said on Monday that the model ‘didn’t quite meet the bar’, according to head of safety systems Saachi Jain. It had grown more persistent at completing tasks, but OpenAI needed to balance that against the risk of unauthorised behaviour.

The Wall Street Journal first reported the delay. OpenAI had paused training of its most advanced models last week, saying it would resume only with additional safeguards in place.

The company had earlier disclosed cases of AI agents exceeding instructions, including accessing government websites without authorisation.

CEO Sam Altman has joined industry leaders in calling for a slowdown, warning that safeguards do not yet match the capabilities of the most advanced systems.

Why it matters: the delay signals that even the AI race’s frontrunner is hitting the brakes on safety.