More than 100 Fortune 500 companies and institutions have regained access under a closely controlled framework, but critics warn that the government’s opaque selection process raises serious rule-of-law concerns.
Anthropic’s most powerful AI model is back online, but only for a carefully chosen few. The US government has partially reversed a two-week-old ban on Anthropic’s Claude Mythos 5, allowing more than 100 trusted organisations, including many Fortune 500 companies, to regain access to the model, which was abruptly disabled over national security concerns on June 12.
What Changed
A letter from US Commerce Secretary Howard Lutnick to Anthropic confirmed that the company had made what he described as significant progress in addressing risks associated with its frontier models. Anthropic said the government specifically cleared Mythos 5, its strongest cybersecurity-focused model, for redeployment to organisations that operate and defend critical infrastructure. Non-US citizen employees of approved companies will no longer require an export licence to access the model. Those outside the approved list, however, remain locked out.
The Controversy Around Access
The government’s hand-picking of approved organisations has drawn swift and pointed criticism. John Coleman of the Foundation for Individual Rights and Expression argued that no one knows how these companies are selected or why others are excluded, calling the process an excessive concentration of power with little transparency. OpenAI CEO Sam Altman echoed those concerns, writing that extensive safety testing is reasonable but that governments choosing customers is not.
Where Fable 5 Stands
Anthropic’s other restricted model, Fable 5, remains unavailable for general use. The government is reportedly moving toward allowing its release, though no timeline has been set. Unlike Mythos 5, Fable 5 is designed for broad public deployment, making its continued restriction particularly consequential for the wider developer community.
The Bigger Stakes
Analysts warn that prolonged restrictions on frontier model releases could hand China a strategic advantage. The longer there is no clear system allowing US companies to widely release updated models, the more ground Beijing can gain, a concern that is quietly shaping the urgency behind Washington’s negotiations with Anthropic and OpenAI alike.
