On June 30, local time, Anthropic announced that the U.S. Department of Commerce had revoked the export restrictions on Mithos5 and Fable5.
In a post on X (formerly Twitter), Anthropic stated, "We will begin restoring access tomorrow" and expressed gratitude to users for their patience and to everyone who worked to facilitate the model's redistribution.
According to a letter from U.S. Secretary of Commerce Gina Raimondo, the department informed Anthropic that no separate authorization is required for the export of the Mithos and Fable models.
Raimondo explained that Anthropic has agreed to proactively detect and respond to security risks associated with the models, consult with the U.S. government on future model release standards and procedures, and notify the government of any malicious activities detected.
Earlier, Politico reported that the Trump administration was expected to lift the export restrictions on Anthropic's Fable5 and Mithos5 as early as that evening. The publication noted that this decision allows Anthropic to offer Fable5 to both U.S. and international users, aiming to ease the controversy surrounding the control of advanced AI models that has persisted for weeks.
Mithos5 is Anthropic's advanced model equipped with sophisticated cybersecurity vulnerability detection capabilities. Fable5 is a version of Mithos5 adjusted for general users, reportedly featuring additional safety measures to prevent malicious use.
On June 12, the U.S. Department of Commerce issued export control guidelines preventing Anthropic from providing access to Mithos5 and Fable5 due to national security concerns. This decision was based on the assessment that military and intelligence agency users in countries of concern, such as China and Russia, could misuse advanced AI models.
Following the guidelines, Anthropic deactivated both models. Subsequently, the U.S. government allowed limited use of Mithos5 for certain "trusted" American institutions, easing some restrictions.
Politico characterized this move as an attempt to balance the need to control cybersecurity risks associated with advanced AI models while addressing pressures not to fall behind in the AI competition with China.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.