OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm

OpenAI said Friday that it is holding back the broader release of its newest family of models, GPT-5.6, after federal officials asked the company to limit who gets access. Rather than opening the models to the public right away, OpenAI is confining early use to a select set of partners it describes as vetted and trusted.

The GPT-5.6 family includes three distinct models built for different needs. Sol sits at the top as the flagship, offering the most capability of the three. Terra is pitched as a middle-ground option suited to everyday tasks, while Luna is designed to be quicker and cheaper to run for lighter workloads.

Despite the differences in power between them, all three models were swept into the same restriction. The Trump administration's request applied uniformly, meaning even the lower-capability Luna and mid-tier Terra are being held back alongside the more advanced Sol. OpenAI said access during this preview period is confined to organizations whose involvement has already been disclosed to government officials.

OpenAI used its Friday statement to push back on the idea that this kind of arrangement should continue indefinitely. The company said it does not want government-directed access controls to turn into a standing practice going forward. It warned that such a pattern would cut off developers, businesses, cybersecurity teams, and international partners from tools they rely on.

The move fits into a broader pattern of Washington pressuring AI developers to rein in their most capable systems. Regulators and officials have increasingly asked leading labs to slow or narrow distribution of cutting-edge models, citing concerns tied to national security and control over powerful technology as it spreads globally.

OpenAI is not the only company to face this kind of intervention. When Anthropic rolled out Fable 5, its top-tier public release, the administration directed the company to cut off access for users outside the United States. Anthropic ultimately went further than a partial rollback, pulling the model offline entirely rather than maintaining a limited version.

The pattern raises questions about how AI companies will balance rapid model releases with government oversight going forward. Industry watchers note the tension between security-driven caution and the competitive pressure companies face to ship advanced tools quickly, especially as rivals abroad continue to advance their own systems. It remains unclear how long the current restrictions on GPT-5.6 will last or what conditions would lead OpenAI to widen access.

For now, the episode signals that government involvement in the release schedules of frontier AI models is becoming more direct, even as the companies building those systems publicly resist treating it as a permanent arrangement. How this dynamic evolves may shape not just OpenAI and Anthropic, but the pace at which the broader industry brings its most powerful models to market.

Source: TechCrunch