OpenAI's GPT 5.6 rollout shows Washington is becoming AI's gatekeeper

Why it matters

The White House does not need a formal AI licensing regime if frontier labs believe broad model access now depends on prior government comfort.

OpenAI CEO Sam Altman told staff on Wednesday, June 24, that OpenAI would release GPT 5.6 first as a limited preview to a small group of partners because the federal government asked OpenAI to stagger the rollout, The Information reported on Thursday.

The report, by Leo Schwartz (@leomschwartz), Stephanie Palazzolo (@steph_palazzolo), and Amir Efrati (@amir), puts OpenAI's next model release inside the same regulatory turbulence that forced Anthropic to pull Fable 5 and Mythos 5 from the market on June 12. Schwartz wrote in a thread on X that the White House clash with Anthropic has created a new operating reality for AI labs preparing frontier-model launches.

According to The Information, Altman told employees during a staff Q&A that the staged release was the fastest path toward broad access. In a Thursday memo, Altman said the government would be "approving access customer by customer during this preview period" for GPT 5.6, according to the report. That is the part of the story that matters: the June 2 White House framework was written as voluntary pre-release cooperation, not a formal licensing regime, but OpenAI is now reportedly treating government approval as a gating function for real customer access.

Altman is not a bystander in that shift. He has spent years arguing that the most powerful AI systems will need a federal oversight mechanism. CSIS noted that Altman told Congress in 2023 that the US should form a new agency able to issue, deny, and revoke licenses for frontier AI development. The Trump administration's June 2 order stopped short of that. It created a voluntary process under which developers of covered frontier models can provide the government up to 30 days of pre-release access, while avoiding mandatory licensing or preclearance.

That distinction is now less clean in practice. The Information reported that a June 9 feedback session hosted by the Office of the National Cyber Director included OpenAI, Meta, and Reflection AI, but not Anthropic. Anthropic was not invited, according to the report. Days later, the White House imposed export controls on Anthropic's most advanced models.

Anthropic's June 12 statement said the US government, citing national security authorities, issued an export-control directive suspending access to Fable 5 and Mythos 5 by any foreign national, including foreign-national Anthropic employees inside the United States. Anthropic said the directive forced Anthropic to disable both models for all customers to ensure compliance. Anthropic also said the government did not provide specific details of the national security concern and that Anthropic understood the issue to involve a possible jailbreak of Fable 5.

Anthropic's response was unusually direct for a company that has built its public identity around AI safety. Anthropic said it had worked with the US government, the UK AI Security Institute, private third parties, and internal teams to red-team Fable's safeguards for thousands of hours before launch. Anthropic also said the capability demonstrated in the reported issue was available from other publicly available models, including OpenAI's GPT-5.5, and argued that applying the same standard across the industry would "essentially halt all new model deployments for all frontier model providers."

The dispute cuts into the founding story of Anthropic. Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and other former OpenAI researchers who had worked on systems including GPT-3, scaling laws, interpretability, and learning from human feedback. Its original pitch was not that it would move faster than OpenAI, but that it would build more reliable, steerable, interpretable AI systems. The irony of the current fight is that the lab most identified with caution has become the test case for the government's most aggressive intervention into model access.

The Associated Press reported this week that an official said testing tied to Anthropic's Project Glasswing found vulnerabilities in classified US systems within hours. The AP also reported that Sen. Mark Warner had referenced the testing at a June 11 Senate hearing, while a group of cybersecurity executives urged the administration to lift the directive, arguing that Mythos models were strong cyber-defense tools but not uniquely capable.

The OpenAI move shows how quickly the Anthropic action has become a precedent. The government did not need to publish a model-licensing rule to change launch behavior. It needed to show that access to a deployed model could be restricted after launch through national-security authorities. Once that happened, every frontier lab gained an incentive to secure government alignment before turning on broad access.

A CSIS analysis put the industry risk plainly: although the Anthropic export control applies only to Anthropic, the confusion around the Bureau of Industry and Security's authority creates uncertainty for the entire US AI industry. CSIS also argued that export controls are not a clear fit, by themselves, for concerns about how AI models are accessed and used, as opposed to the hardware and compute supply chain.

For OpenAI, the immediate calculation is straightforward. A small partner preview lets Altman keep GPT 5.6 moving while reducing the risk that OpenAI becomes the next White House target. For Anthropic, the calculation is harder. Anthropic already tried a restricted partner path with Mythos in April, according to The Information, and still ended up under export controls after Fable's wider release.

That is the new frontier-model launch playbook taking shape in Washington: pre-release government review, selected early customers, and access decisions made one customer at a time. It is not yet a licensing regime on paper. For companies shipping models with cyber capability, it is beginning to function like one.