The Trump administration has asked OpenAI to limit the release of GPT-5.6 to select government-approved partners before any wider launch – citing security concerns stemming from the model’s capabilities, The Information reports.
This is the third major frontier model to face government pre-approval before reaching the public. First it was Anthropic’s Fable 5. Then Google DeepMind’s Mythos 5. Now, GPT-5.6. The pattern is no longer a coincidence – it is policy.
What happened
Under the staggered rollout plan, OpenAI will preview GPT-5.6 to a small group of government-approved partners, with access approved “customer by customer.” The intervention came because GPT-5.6 is considered to have the same capability threshold as Mythos, the Google DeepMind model that triggered its own government review earlier this year.
In an internal memo to employees, OpenAI CEO Sam Altman described the arrangement as the best path to getting 5.6 released at all. A general release is expected “a couple of weeks later” after the partner preview. Altman also told staff that OpenAI has made it clear to the White House that this is not the preferred long-term model for release, and that the company will work toward a more sustainable approach.
Why this matters
The U.S. government is now actively deciding when and how frontier AI reaches the public. The framing is security, not restriction, but the precedent is now well and truly set. As models approach what the government considers the “Mythos-level” capability threshold, sign-off from Washington may become a standard step before wider release.
This creates several dynamics worth watching:
- The bar is moving. If every model that approaches Mythos-level faces pre-release review, the government effectively becomes a gatekeeper for frontier AI capability – whether or not that was the stated intention.
- Competitive asymmetry. Different companies face different timelines. A government review that takes weeks for one lab might take months for another. The playing field is not level.
- Open vs closed. Models released openly or via API may face different scrutiny than those deployed to approved partners. This could accelerate the shift toward controlled, partner-only releases as the default for frontier models.
- Security as the frame. The administration’s framing is national security, not restriction or censorship. That framing is hard to argue against in public debate, which means the precedent may expand quietly over time.
The Mythos threshold
“Mythos-level” has become a shorthand in Washington and Silicon Valley for models that reach a certain capability frontier. Google DeepMind’s Mythos 5 was the first to trigger pre-release government review. GPT-5.6 is now being held to the same standard. Anthropic’s Fable 5 also faced government sign-off requirements.
What exactly constitutes “Mythos-level” capabilities has not been publicly defined in detail, but the consistent pattern across all three cases suggests the government is applying a capability threshold based on internal assessments rather than public benchmarks.
What comes next
Altman’s memo indicated that OpenAI is working toward a “more sustainable release approach” for future models. What that looks like in practice is unclear. The company is balancing between government compliance, competitive pressure, and the expectations of its user base.
For now, the path forward for GPT-5.6 is a staggered partner rollout with government oversight, a general release weeks later, and a long-term question that neither the administration nor the industry has fully answered: who decides when a model is safe enough to release?