Cybersecurity CEO disputes jailbreak label in Anthropic export fight
U.S. limits on Anthropic’s Fable 5 and Mythos 5 models have raised concerns about foreign access rules and AI labs’ ties with Washington.
By Sofia Marchetti · World Affairs Correspondent
3 min read
The Commerce Department used national-security export controls late Friday to block Anthropic from providing its Fable 5 and Mythos 5 AI models to foreign users, Fortune reported. The move matters because Anthropic said the order was broad enough that it disabled the models for all users.
According to Anthropic, the directive covers people outside the United States and foreign nationals inside the country, including non-U.S. citizens who work for the company. Anthropic said it was told the action stemmed from research described as a jailbreak of its systems, involving efforts to test whether safeguards could be bypassed.
Katie Moussouris, chief executive of Luta Security, challenged that characterization after reviewing the findings, according to the Wall Street Journal and her post on X. Moussouris said the work was produced by Amazon researchers and used prompts to draw out information about security vulnerabilities.
In her X post, Moussouris said she had seen the paper and described it as “Defense Oriented Prompting,” or DOP, which she said reflects capabilities needed by defenders. She added that if national defense was the aim, the government action worked against that goal.
Anthropic argued in a blog post that a limited potential jailbreak should not be treated as grounds to pull back a commercial model used by hundreds of millions of people. The company said applying that standard across the AI industry would, in its view, effectively stop deployments of new frontier models.
The company said it supports the government’s authority to restrict unsafe AI systems, but said such decisions should follow a legal process that is transparent, fair and grounded in technical evidence. Anthropic said the Commerce Department’s action did not meet those principles.
Fortune reported that Amazon and the Commerce Department did not immediately respond to requests for comment.
Access rules draw wider concern
The order’s application to foreign nationals in the United States drew criticism from Ben Murphy, a scholar at the Institute for Progress. In a post on X, Murphy described the directive as another move toward splintering technology access by nationality.
Murphy said proof-of-citizenship requirements for services once may have seemed hard to imagine, but have become more common across emerging technologies. He said the latest effort was therefore not unexpected.
Murphy also warned that unpredictable government action could change how AI developers release models. He said labs may keep more systems internal or avoid making them available if they fear sudden restrictions.
According to Murphy, the episode could also make AI companies less willing to share information about potential vulnerabilities with the government. He said Anthropic’s attempt to be open with officials appeared to have hurt the company.
Fortune reported that Anthropic already had a strained relationship with the administration, which has labeled the company a supply-chain risk for Pentagon contractors. Even so, Anthropic gave the government early access to Mythos while warning about possible cybersecurity implications, according to Fortune.
Murphy wrote that he did not know whether the government would have reached the same conclusion without Anthropic’s warnings. He said the public positions involved had not helped create a productive working relationship between the company and the government.
This story draws on original reporting from Fortune.