White House Directs OpenAI to Restrict GPT-5.6 Rollout Pending Security Review

The Trump administration has asked OpenAI to limit public access to its newly released GPT-5.6 model suite, confining the initial rollout to a curated group of government-approved partners while the administration constructs a formal framework for evaluating advanced AI systems. According to Axios, this marks the first time the U.S. government has preemptively asked an American AI company to restrict the launch of a model before its public release. The directive originated from two White House offices: the Office of the National Cyber Director and the Office of Science and Technology Policy. Under the arrangement, OpenAI will preview the model to a select group of partners, with the government approving access on a customer-by-customer basis.

The GPT-5.6 lineup comprises three models: Sol, a flagship model; Terra, a more balanced option for everyday use; and Luna, a faster, lower-cost alternative. Although Sol is the most capable of the three, the Trump administration has restricted the release of all three, with the preview limited to partners whose participation has been shared with the government. A source familiar with the situation told Axios that the government intervened specifically because GPT-5.6 has "Mythos-like" capability, not because the administration is broadly tightening its stance, adding that this level of scrutiny is simply "what's happening with models of that caliber" and that the administration wants to ensure adequate safeguards are in place. OpenAI CEO Sam Altman discussed the model with Commerce Secretary Howard Lutnick the day before the public announcement; Lutnick sought confirmation that all relevant parts of the government had tested and approved the model.

The restriction operates against the backdrop of a June 2, 2026, executive order. President Trump signed an order titled "Promoting Advanced Artificial Intelligence Innovation and Security," which established a process for frontier AI developers to voluntarily provide the government with access to their most powerful models for up to 30 days before public release. That framework has not yet been operationalized: no benchmarks have been finalized for determining which models qualify as covered frontier models, and no formal submission process exists. CNN reported that while Trump signed the executive order earlier this month asking AI companies to submit advanced models for government review, the implementing framework has not been established, leaving confusion among AI companies over which agency is directing AI regulation. Political disagreements within the administration over how restrictive and mandatory the program should be delayed the executive order for weeks.

OpenAI has complied while signaling its objections to the arrangement as a long-term model. Altman told employees in an internal memo that the restricted release is "not our preferred long-term model" and that the company would work toward a more sustainable approach for future launches. In a public statement, OpenAI wrote: "We don't believe this kind of government access process should become the long-term default," adding that it "keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." OpenAI characterized the preview as a "short-term step" that will put GPT-5.6 on the path to broader availability in the coming weeks, as the company works with the administration to develop both a cybersecurity executive order framework and a repeatable process for future model releases. Altman framed the delay as temporary, expressing hope that a broader public release could follow within a few weeks if the preview period proceeds smoothly.

The GPT-5.6 restrictions emerge directly from the administration's prior confrontation with Anthropic. The Commerce Department used national security export controls to bar Anthropic from distributing its Fable 5 and underlying Mythos 5 models to foreign nationals, a category that includes non-citizens working inside the United States, including Anthropic employees. The move marked the first time the U.S. government used export controls to halt access to a commercial AI model already widely used by the public. The government stated that a trusted partner, later identified by CNN as Amazon, had found a jailbreak capable of bypassing Fable's guardrails. After negotiations between the two sides collapsed, the Department of Defense declared Anthropic a supply chain risk, a designation historically reserved for foreign adversaries. Anthropic subsequently sued the Trump administration to reverse the blacklisting, and that litigation remains pending. As of Tuesday evening, the Trump administration lifted export controls on Fable 5, restoring customer access after approximately 18 days of restrictions.

The dual episodes have drawn sharp criticism from legal and policy observers. Brad Carson, head of the bipartisan organization Public First, characterized the government's approach as "ad hoc, personalized, opaque, possibly lawless," while acknowledging that government authority to recall dangerous products, including AI models, must be exercised "consistent with transparency and basic fairness." Axios reported that the U.S. government's desired role in regulating and evaluating frontier AI models before release remains unresolved, creating an ad hoc regulatory environment for AI companies broadly. The Trump administration faces an August deadline under the recent executive order to create standardized benchmarks for evaluating the security risks of new AI models. Until those benchmarks are finalized and a formal submission process is established, the government's authority to condition or restrict model releases rests on informal coordination rather than codified regulatory authority, a posture that courts may be asked to evaluate if the Anthropic litigation advances or similar disputes arise.