The China Mail - Anthropic cuts access to AI models over US 'national security' order

Just three days after publicly launching Fable 5, the company said in a blog post that it received a government directive banning all foreign nationals, even ones who work at Anthropic, from accessing Fable 5 and Mythos 5 over national security concerns.

"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," it said.

The company said it received the letter at 5:21 pm (2121 GMT) Friday.

Axios reported that the letter came from US Commerce Secretary Howard Lutnick.

The US Commerce officials did not immediately respond to a request from AFP.

The firm said that the letter did not state what specifically concerned the government. However, the firm's "understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking'" the Fable 5 model such that it could aid hacking.

Fable 5, released Tuesday, is a locked-down version of Mythos 5, a cutting-edge AI model that Anthropic has held back from the public amid concerns that it had unprecedented abilities to identify software vulnerabilites -- or holes in code that hackers could exploit

Mythos 5 -- the unrestricted model -- has only been released to select companies.

The European Union, which gained access to Mythos earlier in June after weeks of talks, said the latest development further underlined "Europe's need for technological sovereignty".

"We take note of Anthropic's statement and are assessing," said Thomas Regnier, a spokesman for the European Commission, which this month unveiled measures to slash the 27-nation bloc's dependence on America and Asia for key technologies, including AI.

Anthropic said it had reviewed the "jailbreaking" method at the center of the speculation and the hacking opportunities it exposed, but it does not believe Fable 5 gives hackers capabilities that are not already available through other public models.

The firm said that none of its security testers had found a "universal jailbreak," or a way to bypass it's safeguards against helping hackers.

"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company said.

"If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."

Anthropic has been locked in a legal standoff with the Trump administration for refusing to allow its technology to potentially be used for mass surveillance and autonomous weapons, leading the Pentagon to cut contracts with the company.

D.Wang--ThChM