Washington reauthorizes Anthropic's most powerful AI for restricted use in the US

Anthropic and the US Department of War
Reuters/Dado Ruvic/Illustration
After two weeks of an unprecedented lockdown, the United States government has reauthorized Anthropic's most powerful artificial intelligence (AI) model.
The authorization, however, only applies to a small circle of exclusively American partners, which should allow Washington to regain control of the technology, which is considered strategic for the country.
Do you have any suggestions for reporting? Send to g1
According to Anthropic, the conditional unlocking of Mythos 5 will initially only benefit a restricted group of US "cyber defenders and infrastructure operators". The company said it is working to restore access to them "as quickly as possible."
Foreign partners, particularly state cybersecurity agencies in Europe and Asia, remain denied access. The fate of Fable 5, a mass-market version of the Mythos limited in regards to cybersecurity and biological and chemical attacks, also remains uncertain.
Now on g1
The company, whose relations with the Trump administration have been turbulent for months, said Friday that it remains in talks with the government to "expand access to Mythos 5 and make Fable 5 available again" to the general public.
On June 12, Commerce Secretary Howard Lutnick abruptly forced Anthropic to cut off all foreign citizens' access to these two cutting-edge models, citing security concerns. nationwide after flaws were discovered.
The forced withdrawal of a model with cutting-edge technology by a government, something unprecedented, raised a wave of criticism and questions around the world, reviving the debate about the technological dependence of numerous countries on the United States.
Since then, "Anthropic has been working with the US government to reduce the risks associated with the models in question. These efforts have produced significant advances", wrote Howard Lutnick on Friday in a letter to the company cited by several media outlets.
"We work diligently to ensure that the United States remains the world leader in AI while preserving our security," said Commerce Department spokesperson Benno Kass.
The block had been ordered after notification from Amazon about the vulnerability of Fable 5's security systems.
Clients validated by the government
This conditional reactivation occurs on the same day that Anthropic's main competitor, OpenAI, launched its new model, GPT-5.6, also with restricted access and validated on a client-by-client basis by the US government.
"This isn't exactly the process we consider ideal," said Sam Altman, director of OpenAI, although he said the government "is generally doing a good job in a very difficult situation."
These interventions by the Executive, within a still vague and controversial legal framework, enshrine the change adopted by the Trump administration. The American government was previously dominated by opponents of any regulation of AI, who argued it would harm competition with China.
Under pressure from the unprecedented capabilities of these tools, Trump ended up signing an executive order in early June that establishes a federal review of advanced AI models before they are commercialized.
However, the text stipulates that the review is "voluntary" and non-binding.
Some observers see US government control as a development that is likely to favor open-source, free-to-download, modifiable models like China's DeepSeek, making them more attractive to customers who want to avoid dependencies and restrictions.

Source: G1