What happened
On June 25–26, 2026, the Trump administration (via the Office of the National Cyber Director and in coordination with OpenAI's pre-existing DoD agreement) asked OpenAI to limit the initial public release of its GPT-5.6 model family (Sol, Terra, Luna) to a small group of government-approved 'trusted partners.' OpenAI publicly confirmed compliance on June 26, stating: 'At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly.' OpenAI's own blog post (openai.com/index/previewing-gpt-5-6-sol) and System Card confirm the gating. This is the first time the US government has preemptively requested an American AI company restrict the public launch of a model before general release — distinct from the Anthropic case, which was a post-release emergency export control. OpenAI said it does not believe this process 'should become the long-term default' but agreed as a short-term step. The company stated it expects to make the models broadly available within weeks, conditional on the ongoing testing period.
Why it matters
This marks the emergence of a de facto US government pre-clearance process for frontier AI model releases — a structural shift in how AI products reach market. Unlike the Anthropic action (a reactive emergency order), the GPT-5.6 gating was proactive and negotiated. Combined with the Anthropic precedent, it signals that any frontier model with advanced cybersecurity capabilities may now face mandatory government vetting before public access, even absent a formal regulatory framework. Enterprises and developers relying on access to the latest OpenAI models face unpredictable availability windows, and the lack of formal statutory authority or published criteria creates significant legal and operational uncertainty.
Action needed
Enterprises and developers should: (1) register interest with OpenAI to be considered for trusted-partner preview access; (2) review contractual SLAs that assume timely access to latest model versions; (3) monitor OpenAI's announced 'coming weeks' general availability timeline; (4) assess whether their use cases would qualify under any future government vetting criteria as the informal framework solidifies.