OpenAI Restricts GPT-5.6 Release at US Government's Request

OpenAI Restricts GPT-5.6 Release at US Government’s Request OpenAI’s launch of its new GPT-5.6 AI family has turned into a live test of U.S. government power over “frontier” models, pitting security fears against concerns that Washington is quietly creating a de facto licensing regime for advanced AI.

On June 2, President Donald Trump signed an AI security executive order asking leading labs to “voluntarily” submit powerful models for review up to 30 days before release. That framework gained teeth on June 12, when the administration ordered Anthropic to pull its Fable 5 and Mythos 5 models offline after export-control concerns, forcing the rival lab into months of limbo.

As the White House developed a broader testing and evaluation framework, officials from the Office of the National Cyber Director and the Office of Science and Technology Policy then asked OpenAI to limit the rollout of its next model, GPT‑5.6, to a “small set of government‑approved partners,” marking the first time Washington had preemptively sought to restrict an AI launch. The Financial Times reported that Treasury, Commerce and other offices also backed a staggered release to vet users.

On June 25, multiple outlets detailed CEO Sam Altman’s message to staff that customer access would be “approved … customer by customer” during a preview period, with a broader release hoped for “a couple of weeks later.” The Verge described this as a direct response to administration requests over “potential security concerns,” more lenient than the outright export controls imposed on Anthropic.

Less than 24 hours later, on June 26, OpenAI formally unveiled GPT‑5.6 — three models called Sol, Terra and Luna — but only in “limited preview” to roughly 20 partners whose participation “has been shared with the government.” Axios reported that Washington now treats such systems as products requiring pre‑release review, with GPT‑5.6 said to have “Mythos‑like” capabilities. Business Insider noted the constraints stem partly from OpenAI’s agreement that lets the Pentagon use its models.

From OpenAI’s side, the emphasis is on both capability and safety. The company’s system card describes Sol as the flagship of a three‑model family and says GPT‑5.6 ships with “our most robust” safeguards so far. In a tweet, Altman called Sol “a significant step forward,” with Terra offering “5.5‑level performance at half the price,” but conceded that “at the request of the US government, it is launching today in limi…” The Verge reported that OpenAI devoted much of its announcement to cyber misuse, stressing that Sol is better at finding and fixing vulnerabilities than executing “end‑to‑end attacks.”

Publicly, the company is signaling unease with the precedent even as it complies. TechCrunch quoted OpenAI as saying, “We don’t believe this kind of government access process should become the long‑term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.” Another TechCrunch analysis warned that putting every frontier model through an ad‑hoc approval process risks chilling AI investment and turning Trump’s narrower oversight order into a “de facto involuntary licensing regime.”

Industry reactions are colored by the earlier Anthropic episode. The Next Web framed Sol’s 20‑customer preview as the “first practical test” of Trump’s executive order, contrasting OpenAI’s quiet cooperation with Anthropic’s decision to call earlier shutdown demands “disproportionate.” TechCrunch similarly argued that it’s “not about Anthropic vs. OpenAI anymore,” but about whether labs and regulators can jointly design a release process that addresses genuine cyber risks without locking progress — and competition — behind closed doors.

Continue reading https://foxvector.com/stories/019f0640-b6ad-02cb-71b4-26fe4e87431d