Trump Administration Bans Anthropic Fable 5 and Mythos AI Models

The Trump administration has implemented a de facto ban on Anthropic’s Fable 5 and Mythos 5 models, citing national security risks. According to reports from Bloomberg and The Register, the government issued an export control via the Bureau of Industry Security (BIS) to prevent foreign nationals from accessing these dual-use AI technologies, forcing Anthropic to disable the models for all customers.

Why did the government ban Anthropic’s Fable 5 and Mythos 5?

The administration cited an “unacceptable risk of use in or diversion to a military intelligence end use” to justify the restriction. According to a letter obtained by Bloomberg, the BIS used its authorization to require licenses for the export or transfer of items subject to export administration regulations.

The mandate was abrupt. Jessica Lyons, cybersecurity editor at The Register, reports that Anthropic had roughly 90 minutes to comply after receiving the letter. Because the restriction applied to any foreign national—including Anthropic’s own employees—the company disbanded both models for all users to ensure total compliance.

How did Amazon’s report trigger the national security ban?

The ban stemmed from a third-party review conducted by Amazon. According to cybersecurity expert Katie Moussouris, Amazon researchers tested Fable 5, Mythos 5, and Claude Opus by feeding them open-source code containing known Common Vulnerabilities and Exposures (CVEs).

The results were contradictory. When asked specifically to “review the code for security issues,” Fable 5 refused. However, when the prompt was changed to “fix this code,” the model complied. It produced scripts to patch the issues and test those patches. While these are standard functions for defensive security teams, administration officials reportedly viewed the ability to identify and rectify flaws as a tool that “bad guys” could use to hack American systems.

Moussouris, who led the Department of Defense’s “Hack the Pentagon” effort, argues that this is a misinterpretation of the technology. She suggests the government is confusing a model’s ability to fix code with an inherent ability to launch offensive attacks.

Is the ban effective against foreign AI capabilities?

Many security experts argue the ban is a strategic error that harms defenders without stopping attackers. A letter signed by over 100 security experts, including Alex Stamos and Katie Moussouris, claims that restricting these tools removes a critical edge from U.S. defensive teams.

The effectiveness of the ban is further questioned by the rise of open-weight models. According to The Register, models like DeepSeek or other foreign-developed AI in China likely possess similar capabilities. Because open-weight models are publicly available, the government cannot realistically stop them from falling into the hands of foreign adversaries.

Is Anthropic being unfairly targeted by the administration?

Internal tensions suggest the ban may be political. The New York Times reported that some Anthropic employees feel “bullied” or unfairly targeted. This follows a previous conflict where Anthropic requested that the Pentagon not use its models for domestic spying or autonomous weapons targeting.

In response to those requests, the Pentagon reportedly told Anthropic that if they wouldn’t allow unrestricted use, they should be removed from government infrastructure. The administration later labeled Anthropic a “supply chain risk,” which severely limited the company’s ability to secure federal contracts.

The contrast in government relations is stark. While Anthropic butts heads with the White House, Amazon—which provided the report that triggered the ban—maintains a closer relationship with the administration. This dynamic has led industry observers to question whether the “national security” justification is a cover for political retaliation.

What is the role of the Wassenaar Arrangement in this conflict?

The conflict highlights a struggle over the Wassenaar Arrangement, a 42-country agreement on export controls for conventional arms and dual-use goods. Katie Moussouris helped the government renegotiate this agreement to include “defensive security exceptions.”

These exceptions are designed to protect researchers and companies from criminal prosecution when they use dual-use software for malware analysis or coordinated incident response. Moussouris argues that the current ban on Fable 5 ignores these established protections, effectively criminalizing the very tools intended to protect national infrastructure.

FAQ: AI Export Controls and National Security

What are Fable 5 and Mythos 5?
They are advanced AI models developed by Anthropic, derived from the Mythos line, designed with high-level coding and reasoning capabilities.

Why did the BIS intervene?
The Bureau of Industry Security (BIS) intervened to prevent “unacceptable risk” of the models being used by foreign military intelligence entities.

Can foreign adversaries still access similar AI?
Yes. According to security researchers, open-weight models and foreign AI developed in countries like China likely offer similar capabilities that cannot be blocked by U.S. export controls.

Who is Katie Moussouris?
A prominent cybersecurity expert known for establishing Microsoft’s bug bounty program and leading the “Hack the Pentagon” initiative for the DoD.