US Government Tightens Grip on AI Model Releases – OpenAI and Anthropic Face Same Challenges
After pulling Anthropic's models, the US government now restricts OpenAI's GPT-5.6 release to limited preview, raising concerns about unpredictable regulation that could slow the entire industry.
The US government is increasing its control over the release of artificial intelligence models, putting equal pressure on leading AI companies. Two weeks after the government pulled Anthropic’s Fable and Mythos models, OpenAI’s new model GPT-5.6 appears headed for similar limbo. According to The Information, GPT-5.6 will be released only into a limited preview, with the government approving the release “customer by customer” until a general release can be approved.
While OpenAI CEO Sam Altman reportedly projects the preview could last only a “couple of weeks,” Anthropic’s Mythos has already been in preview for months with no indication of a general release. Even a few weeks in review could significantly limit the economic upside of a costly new system, at a time when AI labs are trying desperately to improve their bottom lines. If the pace of model development slows, it could also chill the ongoing data center buildout and put the entire industry at risk.
Critically, OpenAI and Anthropic are now in the exact same position, facing the same problems and the same potential disaster. Conversations in the tech industry often focus on blaming one side or the other—accusing Anthropic of regulatory capture or OpenAI of cozying up to politicians to ice out a rival. But what is happening now is bigger than that. The cost of implementing a haphazard government approval process for every frontier model is obvious, and there is no fix that helps one lab without helping the others.
The most immediate problem is establishing a release process that makes sense. It is fine for the government to test models before release (as is done for many consumer products), but as Dean Ball, a GMU fellow and soon-to-be OpenAI employee, detailed, it is unclear what safety assurances could satisfy regulators. The US government lacks the expertise and capacity for the kind of testing needed. Moreover, it is not even clear what regulators are trying to protect against, since there has been no effort to articulate the specific risks.
Despite these uncertainties, real concerns exist. AI tools are already revolutionizing cybersecurity, with similar processes at work in biorisk and alignment. Restricting model releases alone cannot be the answer—it will only limit what is available to the public. The best ideas for addressing these concerns, as laid out by Ball, involve working together, trusting independent groups to guide the process, and supporting the least-bad regulatory options instead of fighting every regulation tooth-and-nail.
Most importantly, the industry must unite to fight for AI as a whole, rather than seeing safety and regulation as opportunities to gain advantage. AI models have progressed to the point where their capabilities have real political consequences. Dealing with those consequences will require collective action. In the weeks to come, we will find out if the industry is capable of that.
