Anthropic found itself in regulatory limbo this week after the Trump administration issued an unexpected order requiring the company to immediately revoke access to its newest AI models—Fable 5 and Mythos 5—for all foreign nationals. The directive was sweeping: it applied not only to users outside the United States but also to international employees within the company itself. Anthropic, which has significant operations and talent distributed globally, was forced to take its latest capabilities offline while scrambling to understand the legal basis and scope of the restrictions. The company has spent the better part of a week fighting to restore service, but the episode has exposed a critical vulnerability in how frontier AI companies operate under an increasingly fragmented regulatory landscape. Unlike traditional export controls that target specific technologies or nations, these orders appear to apply blanket restrictions to cutting-edge AI models without clear statutory authority or public explanation.
The opacity of the export rules has created immediate practical and legal headaches. Anthropic and other AI developers now face an impossible compliance matrix: which models count as restricted? Do newer iterations automatically trigger controls? How long do restrictions last? Who decides? These questions lack clear answers, and the regulatory environment appears to shift without advance notice. The directive affects not just international users but the global AI research community, which relies on access to frontier models for safety testing, alignment research, and competitive development. By restricting access to Anthropic's latest systems, the administration has effectively prevented independent researchers from auditing these models for risks—the very scrutiny that might demonstrate safety and build public confidence. This creates a perverse incentive structure where companies keep their most advanced systems behind closed doors, making comprehensive safety evaluation nearly impossible.
The timing is notable: this week's Pew Research poll found that 63 percent of Americans believe AI is advancing too quickly, even as chatbot adoption has surged from 33 percent of the population in 2024 to 49 percent today. Yet rather than reassuring the public through transparency and competitive access to advanced models, export restrictions may deepen the perception that AI development is opaque and unaccountable. The hardware acceleration of AI capabilities—evident in new products like Midjourney's ultrasound scanner and Snap's $2,195 AI glasses—suggests the technology is becoming embedded across industries faster than policy can adapt. As AI companies face mounting regulatory pressure with unclear rules, the industry may fragment into a patchwork of regional systems, slowing beneficial innovation while creating security risks through balkanization. Without clearer statutory guidance and public explanation, export controls risk becoming a tool for competitive advantage rather than genuine national security protection.