OpenAI Pulls New AI Model Over Safety Concerns, but There's a Catch

OpenAI scrapped plans to release GPT-6.1 Astra after the model failed to meet the company's safety standards. The company had been preparing Astra as its next model release, but internal testing ended those plans.

OpenAI says the problem was not whether Astra could perform. It was whether the company was comfortable putting the model in front of users. Saachi Jain, OpenAI's head of safety systems, described the release standard as an “extremely high bar” for safety and alignment.

“Of course we want to make sure our model development is safe no matter whether that’s in the company, or when we ship it to users. But when we ship it to users, we have an extremely high bar in terms of safety and alignment.”

OpenAI's decision also came a day before its annual developers conference. The company released GPT-6 Astra earlier this month, but the next version won't be joining it. CEO Sam Altman has meanwhile backed Anthropic's call for the industry to slow down rather than race ahead with increasingly capable models.

President Donald Trump will get his own turn at the issue Tuesday when he headlines the “Golden Age” event at the Andrew W. Mellon Auditorium in Washington, D.C. AI, energy, and space are on the schedule. Elon Musk, Nvidia CEO Jensen Huang, Anthropic co-founder Tom Brown, and Palantir Chief Technology Officer Shyam Sankar are among those listed on the invitation.

Trump has resisted calls to slow American AI development over concerns that China would benefit. Huang has also questioned some of the warnings coming from the safety camp. OpenAI and Anthropic are taking the other concern seriously enough to publicly discuss slowing development.

Now OpenAI has gone further and actually killed a release. GPT-6.1 Astra failed the company's internal standard and stayed behind closed doors. For all the warnings about what powerful AI could do once released, this one didn't get that far.

Earlier this month, former Anthropic researcher Jacob Coxon provided a very different entry point into the same debate. Coxon, who had been at Anthropic for roughly four months before resigning, warned that AI companies were "gambling with our lives" and that increasingly powerful systems could eventually threaten humanity. His warning soon reached Capitol Hill.

Sen. Bernie Sanders (I-VT) cited Coxon while calling senators to a private briefing on the “extraordinary dangers that AI poses for humanity.” Sanders has also backed legislation that would pause AI development.

Then some inconvenient details emerged.

A professional communications firm with ties to the AI-safety movement was working with Coxon as his warning exploded across the media. That came despite Coxon telling Fox News' Bret Baier that he had not worked with third parties on his whistleblowing and public rollout.

That did not prove Coxon's underlying predictions wrong. Nobody has a crystal ball for where this technology ends up.

Astra at least gives Washington something that can be judged by what actually happened. OpenAI put the model through its tests, didn't clear it, and walked away from the planned release. The public never got access.

There are still obvious questions about what happens when a company gets that judgment wrong or doesn't have the same standards. But Astra isn't an example of that. In this case, OpenAI left a product on the table rather than ship a model its own safety team wouldn't clear.

Trump has already taken a more limited approach. His administration established a voluntary process allowing companies to submit powerful AI models to the federal government for review before public release, while the president has continued warning against regulations that could hobble American developers in their competition with China.

China will be part of Tuesday's calculation whether the safety crowd likes it or not. Washington can write rules for American companies. It can't write Beijing's rules for it. Any decision to slow U.S. development has to be made with that difference in mind.

That leaves an odd backdrop for Tuesday. Some of the industry's biggest players will gather in Washington to argue over whether AI is moving too fast just after OpenAI decided one of its own models wasn't ready and stopped it.

The safety concern was real enough for OpenAI to sacrifice a product over it. The product is still sitting inside OpenAI.

Editor's Note: Do you enjoy RedState's conservative reporting exposing the radical left and woke media? Support our work so we can keep bringing you the truth.

Join RedState VIP and use promo code FIGHT to receive 60% off your membership.