AINEWS
A US government order just turned a frontier AI model into a national-security asset
Anthropic disabled global access to its newest model after a directive to lock out foreign nationals, even as the company said it disagreed with the premise.
Anthropic disabled global access to its newest model after a directive to lock out foreign nationals, even as the company said it disagreed with the premise.
The US government has directed Anthropic to suspend access to its newest frontier AI model for all customers worldwide, including the company's own foreign employees, marking what appears to be the first time Washington has moved to restrict a commercially deployed frontier AI model on national security grounds.
Anthropic received the directive at 5:21pm ET on Friday, June 12, according to the company's official statement on the US government directive to suspend access to Fable 5 and Mythos 5, and complied by disabling global access to Fable 5 and a related model, Mythos 5. The company said it disagrees with the premise of the order. The directive orders the lab to lock out any foreign national from the models, including foreign-national Anthropic employees inside or outside the US. Because the only practical way to enforce that constraint is to turn the models off for everyone, Anthropic said, the result is a worldwide suspension of customer access. Engadget reported the access cutoff on June 13.
The story, as told by Anthropic so far, is the first time the US government has treated a commercially released frontier AI model as a national-security-restricted asset. Anthropic said the government verbally shared evidence of a "narrow, non-universal jailbreak," a way to ask the model to read a specific codebase and fix software flaws. The company said the demonstrated capability is widely available from other publicly deployed models, including OpenAI's GPT-5.5, and is used by defenders to keep systems safe. Anthropic called the action a "misunderstanding," said it is complying because the directive is legal, and promised to share more details within 24 hours, as recorded in its official statement. No US agency has been named in available reporting, and no government public statement is on the record; the government's reasoning is currently Anthropic's characterization, not an independent finding.
Fable 5 launched on June 9, four days before the cutoff, and was designed to bring the capabilities of Anthropic's previously restricted Mythos model to the public. Mythos itself is Anthropic's state-of-the-art cybersecurity model, available only to partners in the company's Project Glasswing effort to prevent AI-enabled cyberattacks. Engadget's prior coverage of the Fable launch notes that Fable was red-teamed for thousands of hours before launch, with the US government, the UK AI Safety Institute, multiple private third parties, and internal Anthropic teams participating, and that no universal jailbreak was found. Fable also requires 30-day customer data retention to enable jailbreak research and mitigation.
Anthropic's posture combines compliance with public dissent. The company said it supports a statutory process that is "transparent, fair, clear, and grounded in technical facts" for the government to block unsafe deployments, but said this particular action does not adhere to those principles. Other Anthropic models, including the Claude line, are not affected and remain available. The company said it is "working to restore access" to Fable 5 and Mythos 5.
What to watch: whether a US agency puts its reasoning on the record, anonymously or on the record, by the time Anthropic's 24-hour window closes; whether enterprise customers, particularly outside the US, ask publicly why a model they had access to on Monday was unreachable on Saturday; and whether other frontier labs read this as a precedent for treating commercial AI deployments as dual-use infrastructure subject to export-style controls. The combination of global compliance and a public "we disagree" from the lab is itself a precedent, independent of how the immediate standoff resolves.