OpenAI is facing a new constraint on how it gets its next model into the world, and it didn’t come from a competitor or a compute shortage. It came from Washington.
This marks the second time in as many months that a federal office has reached directly into a frontier lab’s launch plans instead of waiting to regulate after the fact.
Washington has been pushing for early access to AI models for months
President Donald Trump signed an executive order in early June asking AI companies to voluntarily give the government early access to upcoming models, according to NBC News.
The order set a 30-day preview window, down from a 90-day version floated earlier and then abruptly shelved. Officials described the arrangement as voluntary, not a licensing requirement.
Related: OpenAI just built a chip to cut Nvidia out of one job
That order followed unease inside the administration about Anthropic’s Claude Mythos, a model the company said could uncover software vulnerabilities other systems missed, according to CNBC.
The capability impressed investors and worried security officials in equal measure, and it reset how Washington thinks about frontier launches generally.
OpenAI agreed to approve GPT-5.6 access 1 customer at a time
The Trump administration has now asked OpenAI to stagger the rollout of its newest model, GPT-5.6, rather than release it broadly, The Information reported, a request Reuters later confirmed.
CEO Sam Altman told staff the government would be “approving access customer by customer” during the preview period.
The request came out of talks with two federal offices, the Office of the National Cyber Director and the Office of Science and Technology Policy.
A broader release could follow within weeks. For OpenAI, a launch decision is now partly out of its own hands, a first for the company.
The timing matters because OpenAI confidentially filed for an IPO on June 8, targeting a fourth-quarter listing at a valuation near $852 billion, according to CNBC.
ChatGPT already serves more than 900 million weekly users. A customer-by-customer approval process is a different kind of bottleneck, and any visible slowdown in enterprise adoption will get read against a prospectus that is about to invite scrutiny of OpenAI’s growth rate.
That scrutiny may matter less urgently than it did a week ago. OpenAI is now leaning toward pushing its listing into 2027 rather than rushing a fourth-quarter debut, according to the New York Times.
Choppy markets and a rocky post-IPO swing in SpaceX’s stock have left the company’s advisers wary of tepid demand from retail investors.
A staggered government approval process for GPT-5.6 now gives OpenAI one more reason to wait rather than one more reason to rush.
SOPA Images / Getty Images
Anthropic already lived this exact scenario
Weeks earlier, Anthropic went through a harsher version of the same pressure. The company released Fable 5, a public version of Mythos with safety limits built in, alongside Mythos 5 for vetted partners only, according to CNBC.
Days later, the U.S. Commerce Department ordered Anthropic to cut off both models for any foreign national, including its own non-citizen staff.
Anthropic said the directive, sent by Commerce Secretary Howard Lutnick, came with no detailed explanation, and that the underlying flaw was already discoverable in other public models, including OpenAI’s GPT-5.5, according to Fortune.
More OpenAI:
- OpenAI just built a chip to cut Nvidia out of one job
- OpenAI admits enterprises need better control over AI costs
- OpenAI makes IPO decision amid Anthropic, SpaceX fervor
The company complied anyway, since filtering users by nationality across global cloud platforms wasn’t technically feasible.
The shutdown landed days after Anthropic had confidentially filed for its own IPO at a $965 billion valuation, an unwelcome backdrop heading toward a public listing.
That parallel raises a fairness question nobody has answered. If Washington later clears GPT-5.6 for a full release while Mythos 5 and Fable 5 stay shut out, the contrast will look less like consistent enforcement and more like one lab getting a second chance the other didn’t.
For now, the answer from both sides is the same one investors face too: Wait and see what Washington decides next.
Anthropic wants Congress to set the rules instead of agencies
Anthropic’s objection was never that the government acted, but that no statute defines when an agency can pull a model.
The company has called for “a statutory process that is transparent, fair, clear, and grounded in technical facts,” according to its own statement.
It has proposed an Advanced AI Framework that would give regulators clear legal authority to block dangerous deployments, rather than relying on ad hoc directives, according to Anthropic’s policy blog.
The framework would apply only to companies spending more than $1 billion a year on AI research, narrowing its reach to the handful of labs at the frontier.
The pattern is becoming the policy
Both episodes point to the same shift, and it predates this month. The Pentagon labeled Anthropic a supply chain risk to national security in March, a designation usually reserved for foreign adversaries, and Anthropic is fighting that designation in federal court.
Frontier AI models are increasingly treated less like ordinary commercial software and more like a national security asset, something to be cleared, monitored, and occasionally pulled rather than simply regulated after release.
Two labs, two models, two ad hoc interventions inside a few weeks is no longer a coincidence. It is the default way Washington manages frontier AI launches right now, agency by agency, with no statute defining where the line sits.
Both companies are walking toward Wall Street with strategies built for a fast, borderless rollout, and Washington is pricing a slower, supervised one instead.
The market that prices their IPOs later this year will be the one finally forced to decide which version of AI deployment it believes.
Related: White House latest verdict flips script on Anthropic