The U.S. Department of Commerce has lifted export controls that had been applied to Anthropic’s Fable and Mythos artificial intelligence models, the company announced on Tuesday. The reversal comes roughly three weeks after the department restricted access to the firm’s most advanced models over national-security concerns.
What happened and when
Anthropic said the Commerce Department officially withdrew the restrictions, and the company will begin restoring access for affected users the day after the decision. The original measures were introduced under a June 12 decree that limited availability of the Mythos 5 and Fable 5 models. Last week the government had already authorized a release of Mythos 5, but then only to a small number of U.S. organizations deemed “trusted.”
Government conditions and company commitments
Howard Lutnick, the U.S. Department of Commerce’s commerce minister, confirmed the withdrawal in a letter to Anthropic, while reserving the department’s right to reassess the decision if circumstances change or if the company fails to meet its commitments.
Anthropic pledged to proactively detect and manage security risks associated with its models, to cooperate with the government on developing protocols, and to report suspected malicious activity.
Technical fixes and defensive measures
The firm said it has deployed a new defensive measure specifically designed to block a technique the government identified as enabling the “bypass” or “hack” of Fable 5. An unnamed company source told reporters that the vulnerabilities revealed by that technique had already been identified and remediated earlier.
Although Fable and Mythos are built on the same base model, they target different users: Fable is intended for broad public use, while certain security restrictions around Mythos have been eased.
Expert concerns and criticism
Security experts warn that Mythos-level models in the wrong hands could significantly accelerate sophisticated cyberattacks, especially in industries that rely on decades-old, complex systems such as banking. The government’s actions have drawn criticism from market participants who said the process for designating which companies are “trusted” lacked transparency and appeared unfair.
Regulatory context
The tighter controls began with an executive order from President Donald Trump establishing a voluntary framework under which developers can offer affected high-end models to the U.S. government for testing for up to 30 days before making them available to trusted partners.
Separately, both Anthropic and OpenAI have confidentially filed paperwork related to planned initial public offerings (IPOs).
Why it matters
The imposition and subsequent withdrawal of export restrictions on advanced language models highlights the tension between national-security concerns, economic competition and technological innovation. The government’s decision and Anthropic’s security commitments will shape how and under what conditions broader access to these models is restored.
Source: Reuters
