OpenAI Group PBC will roll out its next model, GPT-5.6, to a small group of partners rather than the public at the request of the Trump administration, the latest sign that Washington now wants to review frontier artificial intelligence models before they ship widely.

The plan was disclosed by Chief Executive Sam Altman during a staff question-and-answer session on Wednesday, as detailed in a report from The Information. Altman reportedly told employees the federal government had asked the company to take the staggered approach and that it was the fastest path to a broad release. In a memo, according to the report, he said the government would be “approving access customer by customer during this preview period,” adding that he hoped a wider rollout would follow “a couple of weeks later” if the review went well.

The arrangement mirrors how rival Anthropic PBC handled Mythos, a model with advanced cybersecurity capabilities that it shared with select partners in April rather than releasing it publicly. The Trump administration later forced Anthropic to pull Mythos and a companion model offline earlier this month under an emergency export control directive citing national security, a confrontation that appears to have set the template for how the government now approaches cutting-edge releases.

Reuters also reported the GPT-5.6 review and said the request grew out of talks with two agencies, the Office of the National Cyber Director and the Office of Science and Technology Policy. What the customer-by-customer vetting involves and how long it runs, remains unclear. OpenAI and the White House have not yet publicly commented on the reports.

The episode points to a shift in who decides when powerful models reach the market. AI developers have spent years arguing over how openly to release their systems. Increasingly that call is being shaped in Washington, where officials worry that models capable of finding software vulnerabilities or breaking into hardened systems could spread to adversaries before safeguards are in place.

OpenAI is navigating the new scrutiny as it weighs the timing of a stock market debut that would rank as one of the largest in history. The company is leaning toward waiting until 2027 to go public rather than filing this year, the New York Times reported today.

Recent market turbulence is a factor. SpaceX raised $75 billion in an initial public offering on June 11 that jumped on its debut before surrendering most of those gains, a wobble that has made OpenAI wary of weak demand from retail investors. Chief Financial Officer Sarah Friar has pushed for the later date, pointing to the company’s heavy cash burn, its compute commitments and the demands of public reporting, the Times said. Friar has favored a 2027 listing since at least last year, while Altman has leaned toward moving sooner.

OpenAI completed the corporate restructuring needed to go public in October, converting its for-profit arm into the public benefit corporation now known as OpenAI Group PBC. It was reported at the time that the company was weighing a filing as soon as the second half of 2026 at a valuation of up to $1 trillion. The company closed a $122 billion funding round at an $852 billion valuation on March 31.

A spokesperson previously told Reuters that “an IPO is not our focus, so we could not possibly have set a date.”

Photo: OpenAI