Anthropic CEO Dario Amodei spent months lobbying for government oversight of frontier AI models, warning they were too dangerous to release unchecked. On Friday afternoon, the Department of Commerce used his own arguments as the pretext to suspend all access to Anthropic’s newly released Fable 5 and Mythos 5 models for any foreign national worldwide.
The directive, issued by Commerce Secretary Howard Lutnik, cites a national security concern stemming from a jailbreak reportedly discovered by researchers at Amazon. Anthropic contends the vulnerability is minor and already present in competitor models like OpenAI’s GPT-5.5, arguing that “perfect jailbreak resistance is impossible” for any provider.
"The irony is total. For over a year, Anthropic CEO Dario Amodei has warned that frontier AI is too dangerous to release without government oversight. On Friday afternoon, the Department of Commerce took him at his word."
- Nathaniel Whittemore, The AI Daily Brief
The shutdown doesn’t just block exports to hostile nations; it prohibits foreign nationals from accessing the models anywhere, including inside the U.S. This creates an immediate crisis for Anthropic’s internal development. High-profile researchers like Andrej Karpathy, who are not U.S. citizens, are now legally barred from interacting with the technology they helped build.
Industry reaction is largely unsympathetic. AI builders like Sarah Hooker and Jeremy Howard argue Anthropic’s arrogance and fear-mongering created the nightmare scenario it now protests. Entrepreneur John Enis claimed the company’s safety rhetoric was a strategy for regulatory capture and to mask compute scarcity for serving Mythos.
"By framing AI as a weapon of mass destruction, dominant firms encourage the licensing regimes that stifle open-source rivals."
- Marty Bent, Rabbit Hole Recap
The ban signals a shift from voluntary safety checks to mandatory, arbitrary intervention. If the standard for a recall is a “non-universal jailbreak,” no frontier model is safe from sudden termination. This unpredictability destroys the U.S.’s reputation as a reliable technology provider.
Analysts warn procurement officers in Tokyo, London, and Brussels now have a definitive reason to abandon U.S.-hosted AI. If access can be revoked unilaterally on a weekend, building critical infrastructure on American models is a massive fiduciary risk. The policy effectively subsidizes sovereign AI and Chinese open-weight models like DeepSeek.
CFR senior fellow Chris Magcguire criticized the Department of Commerce’s export control strategy as incoherent, simultaneously allowing advanced AI chips to leak into China while banning model access globally. The long-term impact may be a balkanized global AI landscape, with access to frontier intelligence restricted by citizenship rather than merit.
