OpenAI’s 3 New GPT Models Restricted by Washington—What SaaS Builders Must Know

For SaaS providers building on OpenAI’s APIs, the limited preview of Sol, Terra, and Luna is not just a technical bottleneck but a strategic warning: Washington is now actively curating who gets early access to state-of-the-art models, and that curation will define competitive positioning in the cloud AI market.

On June 27, 2026, OpenAI confirmed it is restricting the initial rollout of its latest GPT-5.6 series—three distinct models named Sol, Terra, and Luna—to a small group of government-vetted partners, a departure from its historical practice of immediate public release. The move, made at the direct request of Washington, marks the first time a flagship ChatGPT iteration has been staggered under federal oversight, signaling a profound shift in the governance of frontier artificial intelligence. Coming less than a month after President Donald Trump’s June 2 executive order directing the Commerce Department to develop a framework for reviewing advanced AI models, the episode establishes an early precedent for White House–led gatekeeping of model launches, blending export control law with ad hoc administrative directives.

“

While Mythos 5 has since been partially restored—Axios and The Verge reported on June 21 that Anthropic had addressed government concerns—Fable 5 remains locked down.

The decision follows closely on the heels of a similar, more abrupt incident at Anthropic. Earlier in June, the company launched its own advanced systems, Fable 5 and Mythos 5, only to disable access three days later after receiving a U.S. government export control directive citing national security concerns. While Mythos 5 has since been partially restored—Axios and The Verge reported on June 21 that Anthropic had addressed government concerns—Fable 5 remains locked down. Commerce Secretary Howard Lutnick underscored the administration’s growing comfort with this model of cooperation in a June 26 letter to Anthropic co-founder Tom Brown, stating that the company’s engagement had “yielded significant progress.”

For OpenAI, the limited preview is framed as a “short-term” bridge while the formal review framework is being built. In its statement, the company emphasized that it had begun “a limited preview of the GPT 5.6 series… for a small group of trusted partners whose participation has been shared with the government,” while insisting that it believes in broad access and intends to make the models generally available “in the coming weeks.” The arrangement underscores a delicate balancing act: maintaining close ties with an administration that has signaled a more interventionist stance on AI, while reassuring enterprise and developer customers that the restrictions are temporary.

The broader context is a dramatic about-face in U.S. AI policy. Upon returning to office in 2025, Trump championed a hands-off, minimal-regulation posture, a stance that fueled breakneck innovation but also attracted criticism from national security hawks. The June 2 executive order—and the enforcement actions against two of the most prominent AI labs within weeks—reflects a rapid pivot. Federal officials now appear to be asserting that frontier models can present export-control risks even when shared domestically, a position that critics argue stretches the Export Control Reform Act beyond its traditional scope.

This shift carries weighty implications for the AI industry. For one, it injects uncertainty into release timelines and partnership negotiations. Foundational model providers may now need to incorporate government review periods into their product roadmaps, potentially adding weeks or months and ceding first-mover advantages to overseas competitors. Already, some observers warn that such ad hoc, non-statutory oversight could reduce predictability for developers and ultimately weaken the competitiveness of U.S. AI firms. The lack of a clear legal framework—with decisions resting on informal requests and letters—raises the specter of regulatory capture and uneven enforcement.

On the other hand, the administration’s approach offers a de facto fast-track to formalizing safety and security evaluations without waiting for Congress to pass comprehensive legislation. The early cooperation of OpenAI and Anthropic may produce a template for future reviews, with vetted partners serving as beta testers for both the technology and the oversight process. Commerce Secretary Lutnick’s positive tone suggests that the government sees these cases as successful experiments, and the partial restoration of Mythos 5 demonstrates a workable off-ramp for companies that comply.