The U.S. Department of Commerce has imposed stringent export controls on Anthropic’s advanced AI models, Claude Mythos 5 and Claude Fable 5. This directive mandates that Anthropic secure an individually validated export license before sharing these models with any foreign national, regardless of location.
Invoking the Export Control Reform Act of 2018 (ECRA), the Bureau of Industry and Security (BIS) emphasized the necessity of these controls to safeguard national security. The order also references the Export Administration Regulations (EAR), highlighting concerns over potential military-intelligence applications of these AI technologies.
The BIS directive encompasses all forms of export, reexport, or in-country transfer of the Claude Mythos 5 and Fable 5 models. This includes:
- Transmitting or transporting the models out of the United States.
- Transferring the models between foreign countries.
- Retransferring the models within a foreign country.
- Releasing the models to foreign nationals within the United States, including non-citizen employees of Anthropic.
Anthropic received this directive on June 12, 2026, shortly after the models’ launch on June 9, and promptly restricted global access to both models.
The impetus for these controls appears to be a reported vulnerability in Fable 5 that could bypass its safety filters, potentially exposing the advanced cybersecurity capabilities inherent in the Mythos 5 framework. Mythos 5 is notably adept at identifying software vulnerabilities, including those in production codebases, and was initially intended for exclusive use by government entities and select enterprise partners. Anthropic asserts that the identified bypass reveals only minor security flaws, comparable to those detectable by other publicly available AI models.
In response, over 80 cybersecurity executives and experts, including leaders from Nvidia and Adobe, have signed an open letter to Commerce Secretary Howard W. Lutnick, urging the administration to reconsider the restrictions. They support Anthropic’s position that the perceived threat is overstated.
To comply with the new regulations, Anthropic must submit license applications through the Simplified Network Application Process Redesign (SNAP-R) portal, referencing the “Is Informed” letter. Non-compliance could result in significant criminal and civil penalties under U.S. law.
Internationally, the European Commission has cautioned that such controls should not be discriminatory, indicating potential diplomatic tensions. Additionally, the restrictions impact foreign nationals on H-1B visas working at U.S. AI firms, raising concerns about workforce dynamics and enterprise access.
This development underscores the growing intersection of AI advancements and national security considerations. As AI models become more sophisticated, governments are increasingly vigilant about their potential applications, especially in sensitive areas. The situation with Anthropic’s models highlights the delicate balance between fostering innovation and ensuring security, a challenge that will likely persist as AI technology continues to evolve.
