U.S. AI Restrictions Create Different Access Paths for OpenAI and Anthropic
Anthropic’s foreign-national suspension and OpenAI’s temporary limited preview have raised questions about UK access to advanced-model safety testing.
U.S. restrictions on advanced artificial-intelligence models have produced different access policies at OpenAI and Anthropic, raising questions about the United Kingdom’s role in independent safety testing.
Anthropic said June 12 that a U.S. government export-control directive suspended access to Claude Fable 5 and Claude Mythos 5 for “any foreign national,” including foreign-national employees working inside the United States. The directive reached Anthropic at 5:21 p.m. ET (21:21 UTC).
The company said the measure did not include detailed national-security information but that it understood the concern to involve a possible “jailbreak” affecting Fable 5. Anthropic said it had not identified a universal jailbreak and described the vulnerabilities it found as minor.
Anthropic also said it had worked with the UK AI Security Institute and other organizations for “thousands of hours” to test Fable’s safeguards before launch. The company offers the same underlying model with different safeguards: Fable for broader use and Mythos for trusted-access cybersecurity and life-sciences work.
Claude Mythos 5.1 was initially available only to selected U.S. organizations. Anthropic said it was coordinating with the U.S. government to expand access to domestic and international partners. It also identified the Life Sciences Verification Program as supporting access to advanced biology capabilities in Mythos 5.1.
OpenAI adopted a narrower, temporary approach. On June 26, the company said the U.S. government had asked it to begin GPT-5.6 Sol, Terra and Luna with a limited preview for a small group of trusted partners whose participation had been shared with the government.
“We don’t believe this kind of government access process should become the long-term default,” OpenAI said. The company also said it was working toward broader availability.
OpenAI later said GPT-5.6 Sol was rolling out globally and that the UK AI Security Institute had received early access for independent, pre-deployment cybersecurity evaluation. That differs from Anthropic’s initial restriction of Mythos 5.1 to selected U.S. organizations and trusted-access programs.
A UK parliamentary question dated Sept. 16 asks the government to assess whether Anthropic decided not to provide Mythos 5.1 to the UK AI Security Institute for pre-release testing. The question is due for an answer by Sept. 30.
