The U.S. government has decided to relax export regulations on Anthropic’s leading artificial intelligence models following an agreement between the company and the Commerce Department. This information was relayed in a letter from Commerce Secretary Howard Lutnick to Tom Brown, one of Anthropic's co-founders, which was obtained by WIRED.
The restrictions are being lifted for both the Fable 5 model and its more advanced counterpart, Mythos 5, which had previously only been available to a limited number of businesses and government entities. Secretary Lutnick stated, “A license is no longer required for the export, reexport, or in-country transfer, including deemed export or deemed reexport, of the Mythos or Fable models.”
These changes come as Anthropic has been collaborating with the Commerce Department and the White House to enhance security measures against the potential misuse of Fable’s features, particularly regarding cybersecurity vulnerabilities, according to insiders familiar with the negotiations.
In the letter, Lutnick noted that Anthropic has committed to actively identifying and mitigating security threats linked to these AI models. The company is also expected to work closely with the U.S. government on developing protocols and standards for Mythos and Fable, as well as any future iterations they might introduce.
Lutnick has been at the forefront of efforts to address the issues with Anthropic, working alongside national cyber director Sean Cairncross. Initially, Anthropic argued that the administration's security worries were exaggerated, claiming it was impossible to guarantee that no escape routes could be utilized to access the advanced functionalities of the Mythos model.
Recently, Anthropic shifted its approach to expedite the return of Fable, leading to changes in how the company interacts with government officials. According to previous reports by WIRED, CEO Dario Amodei has now been replaced in discussions by co-founder Tom Brown, who appears to have made a more positive impression on officials.
Moreover, Anthropic has pledged to enhance security features to minimize the risk of jailbreaks, essentially aligning its message with what the administration sought, rather than debating the core issue of whether jailbreaks can be completely prevented, according to sources.
An update has been added to this report, referencing the letter from the Commerce Department to Anthropic that was reviewed by WIRED.
