OpenAI’s launch of its next-generation GPT-5.6 models has become an early test of how far Washington will go in controlling frontier AI — and how much Silicon Valley is willing to accept in return for staying online.

On June 2, President Donald Trump signed a narrow AI executive order asking leading labs to voluntarily submit advanced systems for up to 30 days of pre-release review, a move critics say has created a “de facto involuntary licensing regime” for powerful models.¹

Government steps in

In mid-June, the administration forced Anthropic to revoke access to its Fable 5 and Mythos 5 models via an export-control-style directive, the first time a commercial frontier AI system was pulled offline by Washington.² Officials argued these systems had “Mythos-like” cyber capabilities that warranted tighter control.³

Against that backdrop, the White House’s Office of the National Cyber Director and Office of Science and Technology Policy asked OpenAI to “limit the rollout of GPT-5.6” to a small set of government‑approved partners while a new security testing framework is built out.³ The Financial Times separately reported that Treasury, Commerce and other offices wanted a staggered release to vet users.⁴

OpenAI agrees to a staggered preview

On June 25, reports surfaced that “The White House is asking OpenAI to slow roll the release of its new model over safety concerns,” with CEO Sam Altman telling staff the government would approve access “customer by customer” during a preview period.⁵ The Verge likewise noted that “the Trump administration… asked OpenAI to stagger the release of its next big-ticket model, GPT-5.6,” with customer access approved case by case.⁶

Axios described the move as “the first time the U.S. government has preemptively asked an American AI company to restrict the launch of a model before release.”³

GPT‑5.6 arrives — but only for a few

Less than 24 hours after those reports, OpenAI formally unveiled GPT‑5.6, a three‑model family — Sol, Terra and Luna — in a “limited preview” closely coordinated with Washington.⁷ Sol is the flagship; Terra is a mid‑tier model “for high-volume work,” and Luna is a “fast and affordable” option, all positioned as particularly strong at coding and cybersecurity.⁷

A safety-focused system card described these as “our new flagship model… a capable lower-cost option; and… our fastest and most cost-efficient model,” emphasizing “the safeguards we have built for this launch—our most robust yet.”⁸

Business Insider reported that, under an existing agreement that lets the Pentagon use its models, OpenAI previewed GPT‑5.6 for the U.S. government and then, “at their request,” started “with a limited preview for a small group of trusted partners whose participation has been shared with the government.”⁹ Axios said that amounted to “around 20 companies, whose participation has been approved by the government.”¹⁰ The Next Web called it “the first time an American AI company has launched a frontier model under a government-managed access list.”¹¹

OpenAI and industry reactions

In public statements and blog excerpts reported by multiple outlets, OpenAI stressed that it does not want this model of government gatekeeping to become permanent. TechCrunch quoted the company saying, “We don’t believe this kind of government access process should become the long-term default,” warning that it “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”¹²

Axios similarly cited OpenAI’s blog: “We don’t believe this kind of government access process should become the long-term default… We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the administration to develop the cyber Executive Order framework and a repeatable process for future models.”¹⁰

Altman echoed the trade‑off directly on X, calling Sol “a smart, efficient, and a significant step forward” but adding: “Bad news: at the request of the US government, it is launching today in limi…” — a reference to the restricted preview.Sam Altman (sama)¹³

OpenAI co‑founder Greg Brockman focused on performance and user interest rather than the politics, writing simply: “GPT-5.6 Sol preview — it’s a good model,” while linking to the company’s announcement.Greg Brockman (gdb)¹⁴

Government and policy perspective

From the administration’s side, officials have framed this as a calibrated response to emerging cyber risks. Axios reported that Commerce Secretary Howard Lutnick wanted “to be sure all relevant parts of the government have tested and approved the model,” with a source insisting “this is what’s happening with models of that caliber,” not a sudden heavy hand.³

Another Axios piece noted that Washington is “starting to treat the most advanced U.S.-developed AI models as products that need government review before they can be widely released.”¹⁰

A follow‑up TechCrunch analysis argued that with both Anthropic and OpenAI now subject to similar constraints, “the U.S. government is set to take an awful lot of control over which AI models get released,” and warned that a clumsy approval regime could “significantly limit the economic upside of a costly new system” and even “put a similar chill on the ongoing data center buildout.”¹

Wider AI community: regulate APIs, not open source?

Beyond the companies directly involved, AI researchers and open‑source advocates have seized on the GPT‑5.6 rollout as evidence that policy needs to distinguish between closed, centralized APIs and freely distributed models.

Hugging Face CEO Clément Delangue argued on X that “It’s quite rational to regulate frontier API models, especially to get more transparency for the government, without regulating open-source AI,” contending that “the most dangerous AI systems right now aren’t open models” but large proprietary APIs.clem 🤗 (ClementDelangue)¹⁵ Meta AI chief scientist Yann LeCun amplified that view by retweeting Delangue’s thread.Yann LeCun (ylecun)¹⁶

In a separate post, Delangue highlighted structural limits on controlling open weights, sharing an interview clip that said: “Open weights are fundamentally different than an API. The way you can restrict it is very different… If you remove an open weight from…” — suggesting takedown-style controls that work for closed services are ill‑suited to decentralized models.clem 🤗 (ClementDelangue)¹⁷

He also contrasted efforts to police open-source with the idea of the government itself training and releasing public models, writing: “Tired: The US government regulating open-source AI models. Wired: The US government training and releasing open-source AI models,” in response to an announcement of a small in‑browser privacy model called Rampart.clem 🤗 (ClementDelangue)¹⁸

What comes next

OpenAI says it expects to expand GPT‑5.6 access “in the coming weeks” if the preview proceeds smoothly, aligning with the administration’s stated goal of building a repeatable process rather than permanently throttling releases.⁹¹⁰

But with Anthropic’s Mythos still stuck in limbo months after its own “preview” began, analysts warn that GPT‑5.6’s staggered debut may be less a one‑off compromise than the model for how frontier AI reaches the public — or doesn’t — in the United States.