White House asks labs to delay UK model tests, Politico says

The White House has asked OpenAI and Anthropic not to give their newest AI models to the U.K.’s AI Security Institute until the U.S. government has tested them first, Politico reported Thursday, citing a person familiar with the matter and a senior administration official, both unnamed. According to the report, the request came from the Office of the National Cyber Director (ONCD).

The ask is about who tests new frontier models first, before release. The senior official told Politico that Washington wants to review the models and make sure U.S. systems are secure before they are shared with partners. “Because they’re American companies and this has been our policy with every new frontier model that comes out,” the official told Politico.

No White House, ONCD, OpenAI, or AISI document or on-record statement confirming the request was available as of Thursday evening. The White House and OpenAI did not respond to Politico; Anthropic declined to comment. Reuters matched the story citing Politico and also got no comment.

The rule in English

The request (as reported). New frontier models from OpenAI and Anthropic would go to U.S. government testing first. Only after that review, and after steps to secure U.S. systems, would the labs share them with the U.K. AISI. Politico describes it as a request, not an order. No legal authority or penalty is reported.

What Anthropic has done (primary). Anthropic launched Claude Mythos 5.1 on Sept. 1. Its announcement says Mythos 5.1 is “only available to a set of US organizations, though we’re coordinating with the US government to expand access to a broader set of domestic and international partners as quickly as possible.” Mythos 5.1 is the same model as the generally available Fable 5.1, with looser cyber and life-sciences safeguards for vetted users. The Financial Times first reported on Sept. 9 that AISI did not get Mythos 5.1 before launch, a first for Anthropic. Politico said Anthropic “appears to have agreed to the request, so far.” Anthropic’s page does not mention the U.K. or ONCD.

What AISI says (via Politico). In a letter to a U.K. parliamentary committee earlier this month, AISI Director Henry de Zoete acknowledged the missing Anthropic access, Politico reported. He wrote that AISI still has pre-release access to “some of the world’s most capable models,” and pointed to its pre-release testing of OpenAI’s GPT-6 Astra.

What London says (via Politico). A U.K. government spokesperson told Politico that AISI “continues to work closely with the U.S. government, international partners and leading AI companies, including OpenAI and Anthropic.”

U.S. capacity. Politico reported that the Commerce Department’s Center for AI Standards and Innovation (CAISI) has no permanent director and only a few dozen technical staff while it is asked to evaluate a growing stream of frontier models. The report does not say which U.S. agency would run the review.

Who wrote it / who it binds

Writer: ONCD, per Politico’s anonymous sources. Binds no one on paper. It works only if the labs comply. Anthropic’s U.S.-only Mythos 5.1 gating is consistent with compliance so far; OpenAI’s response is unknown.

Who it exempts

Other labs (Google DeepMind, xAI, Meta) are not named in the report. General-release models with standard safeguards, such as Fable 5.1, are not described as covered. The report does not say whether EU testers are affected.

What a founder, parent, or worker should do differently

Frontier labs: if the reported request holds, a U.S. government first look becomes a de facto gate on foreign pre-release testing, even without a rule. Plan release calendars around it.

Allied governments and evaluators: pre-release access that used to be voluntary and early would be sequenced behind Washington. AISI’s privileged position is the first thing at risk.

Parents and workers: not a consumer-app rule. It would decide who tests the most capable models before the public gets the safer versions.

What’s next

• On-record confirmation or denial from ONCD, the White House, or OpenAI.
• Whether OpenAI’s next frontier model goes to AISI before or after U.S. review.
• CAISI staffing and director fight in Congress.
• For context, the morning 20-nation call for mandatory pre-deployment testing and independent evaluator access (our lead story this morning) was signed without the U.S. or China.

Source: Anthropic, “Introducing Claude Fable 5.1 and Claude Mythos 5.1,” Sept. 1, 2026.
Secondary (request itself): Politico, Sept. 24, 2026; Reuters match via Devdiscourse. FT Mythos 5.1/AISI report matched via ITPro.