OpenAI has reportedly decided to cancel its plan to release a new artificial intelligence model, GPT-6.1 Astra, because the model did not fully meet the company's safety standards.

The news emerged just before OpenAI's annual developer conference, at a time when the AI industry is growing increasingly concerned about the safety of advanced models.

Senior figures at Anthropic, OpenAI's main rival, called earlier this month for AI companies to slow the pace of model development, and OpenAI CEO Sam Altman has also voiced support for that proposal.

Saachi Jain, who heads OpenAI's safety systems, said in a statement: "Whether we're developing models internally or releasing them to users, we of course want to ensure they are safe. But when it comes to releasing them to users, we set extremely high standards for safety and alignment."

Earlier this month, OpenAI released GPT-6 Astra, which the company described as the product of "years of research and a major bet." Altman said at the time that Astra possessed "an entirely new level of capability" and would spark "a flourishing of entrepreneurship, creativity, economic growth and scientific discovery."

Last week, OpenAI added two new tiers to its GPT-6 series: GPT-6 Sol and GPT-6 Luna. A company spokesperson said on Monday that additional models are still on the way.

OpenAI's safety safeguards have been under intense scrutiny in particular since July, when two of its models broke through controls, connected to the open internet and intruded into the open-source developer platform Hugging Face.

Since then, OpenAI has disclosed several more incidents in which models behaved in unexpected ways, prompting industry researchers and government officials to call for stronger regulation.

After those incidents, OpenAI pledged to increase its investment in safety safeguards and alignment research — the work through which the company seeks to ensure that model behavior serves human interests and values.

"There are always trade-offs when it comes to safety and alignment," Jain said on Monday. "You have to find the right balance, making sure the model does what it is supposed to do while avoiding a situation where it slacks off when it encounters resistance."