The U.S. government will approve access to GPT-5.6 for every customer: an unprecedented solution
Main chat
A chat for vibe coders: news, guides, live cases, marketplace, and finding executors.
On June 25, 2026, the publication The Information, citing internal correspondence OpenAI reported that the new model GPT-5.6 will not be released as usual. Instead of running through the API and ChatGPT for everyone, access at the preview stage will be approved by the US federal government itself - client after client. The news was confirmed within hours by Axios, Reuters, Bloomberg and The Verge that for news of this kind it is a serious indicator of credibility.
Let’s look at exactly what happened, who’s behind it, and why it’s not a one-off episode, but part of a broader pattern that began two weeks ago with a similar story from Anthropic.
What Happened: The Chronology of a Day
On Thursday, June 25, OpenAI CEO Sam Altman told employees at a Q&A meeting that GPT-5.6 will be released in a limited preview for a small group of corporate partners. According to an internal memo cited by The Information, Altman put it this way: The government will "approve client-by-client access during this preview period," and a wider release is expected "in a couple of weeks" after that.
The source of the request is two specific White House entities: the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP). According to Axios, Commerce Secretary Howard Lutnick personally joined the process: on Wednesday, on the eve of the announcement, he discussed GPT-5.6 directly with Altman and insisted that all relevant departments test and approve the model before it sees the light of day.
A source familiar with the situation explained to Axios the logic of the intervention in the most specific way: the government is acting because GPT-5.6 has “Mythos-level capabilities” – a direct reference to the Anthropic model, around which a similar but more severe scandal erupted two weeks ago.
Altman's Quote Showing the Real Balance of Power
In a memo to employees cited by The Information, Altman gave wording that says a lot about the company's position in this situation: "We have made it clear to the U.S. government that this is not the long-term model we would prefer and we will work with them and other industry actors to achieve a more sustainable approach for future releases.".
The wording is notable in that it simultaneously accepts the demand here and now and publicly distances itself from it as a permanent practice. It’s the typical language of a company that doesn’t have real room for rejection, but wants to denote it nonetheless: we see it as a temporary measure, not a new standard of play.
Anthropic link: same logic, two weeks in a row
The context of this story cannot be understood without the events of two weeks ago. On June 12, 2026, the U.S. government issued an export directive obliging Anthropic to completely disable access to Fable 5 and Mythos 5 models — not just for foreign users, but for virtually everyone, including paying customers in the U.S., since the company did not have the technical ability to verify users’ citizenship selectively.
An analyst at one of the publications that tracks the topic puts the parallel very bluntly: the government blocked the Anthropic model on June 12. It now regulates the OpenAI model on June 25. It's two weeks. Both forward camps now operate under the same state release approval process.
In doing so, it is important to capture a key difference that The Verge is highlighting: the conditions for OpenAI are noticeably milder than what happened with Anthropic. Fable 5 and Mythos 5 were completely disabled with no right of access at all – this was a post facto reaction to the already identified vulnerability. GPT-5.6 receives a controlled, phased launch with the prospect of a full release "in a couple of weeks" - a preventive measure agreed in advance, not an emergency shutdown of an already working product.
Voluntary vetting that quickly becomes not very voluntary
Another important layer of history is revealed here – the formal legal framework within which both cases should fit.
Earlier in June, President Trump signed a presidential executive order on AI security that explicitly obliges several federal agencies to build a voluntary testing protocol for AI companies before releasing a new model. The order includes a process by which developers can voluntarily grant early access to the government - up to 30 days before the model's public release.
The word "voluntarily" in this decree is doing less and less work every day. Two events in a row - the shutdown of Fable 5/Mythos 5 on June 12 and the requirement for a phased release of GPT-5.6 on June 25 - make the formally optional framework a de facto mandatory practice with better PR. One commentator on social media directly noted the irony: the decree of June 2 spoke of voluntariness, and two weeks later the government turned off one model and demanded a phased release of another.
However, there is a counter-argument that comes from sources close to the process: according to Axios, OpenAI actually worked with the administration to release GPT-5.6 proactively and in advance – that is, before Anthropic lost access to its models under the compulsory directive. This is more an attempt to build a working dialogue than a passive submission to an ultimatum that came suddenly.
What it means in practice for business and developers
The implications of this decision go beyond a single company model to the predictability of working with advanced systems as infrastructure.
Access through the standard API is not guaranteed right away. The limited preview structure means that GPT-5.6 can technically be considered released but not available through the usual API or ChatGPT interface for most users – access will only go through a federal verification process, which can take weeks, depending on the bandwidth of the relevant agencies.
Competitive inequality between companies with different levels of government ties. Organizations with pre-existing relationships with government structures will logically gain access faster than those who do not – this is a direct consequence of any system of individual approval, regardless of the good intentions of those who build it.
Competitive pressure on OpenAI itself. Unlike OpenAI, Google doesn’t have to wait for a calendar and approvals for its models — at least not in this particular story. If GPT-5.6 spends weeks in state-verified preview mode, any client building security systems, code agents or internal automation on this infrastructure will have to ask the practical question of whether it is possible to build a long-term architecture on an infrastructure whose release schedule can be interrupted by a decision from Washington.
OpenAI on June 8 officially confirmed the confidential filing of Form S-1 with the Securities Commission – that is, the process of going public is parallel to this story, and any instability in the release schedule of flagship models becomes a factor that potential investors will put in the valuation of the company.
Important reservation: what has not yet been confirmed
When dealing with such fast-paced stories, it is useful to clearly fix the line between what is confirmed by reputable publications and what remains the assumption of commentators.
Confirmed by multiple sources (The Information as a primary source, hereinafter Axios, Reuters, Bloomberg, CNBC, The Verge): the very fact of requiring a phased release, the involvement of ONCD and OSTP, Altman’s personal conversation with Lutnick, the wording about the approval of “client by client”, Altman’s quote from an internal memo.
Not officially confirmed by either party: The specific specifications of GPT-5.6 itself, the precise list of criteria by which the government will determine approved customers, and whether this process will apply to OpenAI’s international customers in the same way as it did to foreign users in the case of Anthropic. Neither the White House nor the Office of the National Cyber Leader responded to requests for comment at the time the materials were released.
President Trump has publicly stated that he does not consider Anthropic a national security threat — a statement made after the shutdown of Fable 5 and Mythos 5 — and creates a curious contradiction between the rhetoric and specific actions of the administration, which has not yet received a clear public explanation from either side.
What this means for the industry as a whole
If we consider this story not in isolation, but as part of a sequence, a picture emerges that goes beyond the fate of one particular model.
Until June 2026, testing the safety of advanced models was largely an internal affair of the developers themselves, sometimes supplemented by voluntary cooperation with the state. After June 12 (the shutdown of Fable 5/Mythos 5) and June 25 (the phased release of GPT-5.6 under government control), this process has become a mandatory step in practice for every company purporting to be a leading laboratory in the United States, regardless of whether the state itself uses the word “voluntarily” in formal documents.
The line between “the government tests the model before it is released on the company’s own free will” and “the government decides who gets access to the model” has all but blurred in these two stories. For users and companies building products based on the APIs of leading labs, this means that the availability schedule of the new, most powerful models now depends not only on the technical readiness of the developer, but also on the capacity and priorities of specific federal agencies – a factor that was not included in any planning roadmap a month ago.
*This article is based on The Information, Axios, Reuters, Bloomberg, The Verge, Engadget and Shacknews. Current June 26, 2026; the situation is evolving, the article will be updated as official confirmation of details becomes available. *
