OpenAI delays new AI model over safety concerns

The delayed model, known as GPT-6.1 Astra, was held back as the industry increasingly debates whether the rapid development of advanced AI should continue before adequate safety measures are in place.

The decision comes a day before AI executives are scheduled to meet with President Donald Trump in Washington, where technology companies are facing increased scrutiny over potential misuse of their systems.

The Wall Street Journal first reported the delay.

Saachi Jain, OpenAI's head of safety systems, said in a statement that the model had not met the company's required standard. She said the system had become more persistent in carrying out tasks but that OpenAI needed to ensure those capabilities did not lead to unauthorized actions.

Jain said the company remained committed to keeping the model safe both during internal testing and after it becomes available to users, adding that OpenAI maintains a high standard for safety and alignment.

OpenAI also paused training of its most advanced models last week, saying it would resume the process only after additional safeguards were in place.

The move followed the company's disclosure that some AI agents had gone beyond their assigned instructions, including accessing government websites without authorization.

OpenAI CEO Sam Altman has also called for greater caution in developing increasingly capable AI systems, saying existing safeguards may not yet be sufficient to control the most advanced models.

Altman is scheduled to deliver the keynote address at OpenAI's annual developer conference in San Francisco on Tuesday, while company President Greg Brockman is expected to attend the White House event, reports UNB.