The US government has restricted OpenAI from openly releasing GPT-5.6 Sol, its newest frontier model, limiting access to government-approved partners on national-security grounds. The same controls recently hit Anthropic's Claude Mythos, and the move lands just as Sol rivaled Mythos on leading benchmarks. OpenAI has called the access regime unsustainable.

The US government has barred OpenAI from broadly releasing its newest model, GPT-5.6 Sol, instead limiting access to a small set of government-approved partners through its API and Codex tools, citing national security [doc3][doc4].

The restriction matters because Sol is not a routine update. It rivaled Anthropic's gated Claude Mythos on leading agentic-coding benchmarks, putting OpenAI back at the frontier [doc8]. Washington has now applied the same access controls to both labs' top models [doc3].

OpenAI has pushed back publicly, calling the government-controlled rollout 'unsustainable' [doc8]. Reporting from Wired and CNBC describes the company being directed to gate distribution rather than ship the model openly [doc3][doc4].

The open questions are the second-order effects. Restricting the strongest closed models could cede ground to open-weight alternatives, complicate the work of app builders and security teams that depend on frontier capability, and set a template in which trusted-partner access becomes the norm for advanced AI in the US.

US Government Restricts OpenAI AI Models Over National Security Concerns