GPT 5.6 is being approved customer by customer by Washington during a limited preview. OpenAI says it is complying — and warning that this arrangement is not how it wants to release future models.
The US federal government, not OpenAI, is now choosing which partners get early access to the AI company's newest model. OpenAI's GPT 5.6 is shipping as a limited preview to a small group of partners approved by Washington one customer at a time, according to internal memo coverage first reported by The Information and carried by The Guardian.
The arrangement is the most visible operational test yet of a voluntary US framework for vetting the most powerful AI systems, and OpenAI is publicly signaling it does not expect this to become the long-term playbook. In a memo to staff reported by The Information and summarized by Reuters, CEO Sam Altman wrote that the federal government would "approve access customer by customer" during the preview period, with a broader release targeted "a couple of weeks later" if the process goes well. Altman also told staff, according to the same memo coverage, that OpenAI had "made clear to the US government that this is not our preferred long term model, and will work with them and others in industry to achieve a more sustainable approach for future releases."
The mechanism underneath the request is a voluntary vetting framework executed through two White House policy offices: the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP). The Decoder reports that GPT 5.6 access now requires US government sign-off on a customer-by-customer basis, and Politico has corroborated the regulatory-press framing. The structure echoes how Anthropic, OpenAI's closest rival in the frontier-AI race, launched its Mythos model — also a staged, customer-by-customer rollout.
The pattern has since moved further than OpenAI's case suggests. According to the Guardian's excerpt of the reporting, Anthropic ultimately pulled Mythos from public release after the US government ordered the company to stop foreign nationals from accessing the model. Trade press coverage has framed the US motivation as "safety and security concerns," though the specific concerns have not been detailed in public reporting. What is documented is that the executive branch can move from staged approval to outright withdrawal on a single direction.
Three critique threads sit underneath the arrangement, and any of them could outlast the current White House. First, the vetting framework is voluntary, not statutory: it rests on an executive order and informal pressure, which means a future administration could reverse it without Congress. Second, Politico and the trade press report that the gate is being moved by individual officials, including Commerce Secretary Lutnick's reported push for broader agency approval, rather than by a transparent public process. Third, the Anthropic Mythos precedent — a frontier model pulled entirely after a foreign-national access block — shows that the floor of executive leverage over frontier AI is total withdrawal, and no published ceiling exists.
SiliconAngle has framed OpenAI's compliance in part as a 2027-IPO calculation: the company has little incentive to fight a federal gate when its public market debut is on the horizon. That framing does not contradict Altman's own line that the arrangement is unsustainable for the industry long-term. The two statements can sit side by side: OpenAI is willing to comply for now, and OpenAI believes a more permanent regime, written somewhere other than an executive order, is overdue.
What to watch next is whether the preview expands on the timeline Altman described, whether the federal government publishes the vetting criteria it is using, and whether any rival frontier-model lab — Anthropic, Google DeepMind, or Meta — is asked to follow the same customer-by-customer approval path. The strongest falsifier for the governance-inflection reading is simple: if a future administration reverses the framework without congressional action, the whole arrangement was a single-administration choice, and the leverage disappears with it.