U.S. Orders Anthropic to Limit Export of Fable and Mythos AI Models

U.S. government ordered Anthropic on Friday to limit export of its Fable and Mythos AI models, citing national security concerns. The order did not specify reasons for the restrictions, according to Anthropic.

The company responded by suspending access to both models for all users worldwide. Anthropic had released Mythos in preview in April, initially granting access to about 50 companies before expanding to roughly 150 organizations across 15 countries. Last week the company released Fable, a public version of Mythos that included guardrails to block use in biology, chemistry, and cybersecurity.

Many cybersecurity experts found the guardrails prevented any prompts related to cybersecurity. Anthropic stated the export-control order may have been based on a non-public Amazon research paper describing a method to bypass Fable’s guardrails. Katie Moussouris, one of the signatories of the open letter, reviewed the paper and wrote that it did not demonstrate a real jailbreak.

She said the Amazon researchers asked Fable to fix open-source code containing public and deliberately planted vulnerabilities after the model initially refused to review the code for security issues. Moussouris wrote that the behavior described in the paper cannot be meaningfully fixed without weakening the model for defense.

“Defenders need to be able to ask AI to fix the bugs in a file, explain why the fix matters, and write tests that confirm the patch works,” she stated.

“That is not a guardrail bypass. ” Seventy-six cybersecurity experts signed an open letter urging the White House to lift the export-control order. Signatories include Alex Stamos, Casey Ellis, Jon Callas, Paul Vixie, Dino Dai Zovi, Katie Moussouris, and Rachel Tobac.