Anthropic has disabled global access to its two latest frontier AI models—Fable 5 and Mythos 5—following an expert control decree issued by the U.S. Commerce Department. The agency determined that both models pose a national security risk under current export control frameworks, effectively forcing the company to suspend access worldwide while compliance and licensing pathways are worked out.
The decision has drawn sharp criticism from AI researchers, civil liberties advocates, and industry analysts who argue the move sets a dangerous precedent for government control over AI capabilities.
What Happened
The Commerce Department's Bureau of Industry and Security (BIS) issued an export control determination targeting Fable 5 and Mythos 5, citing the models' advanced capabilities in areas the government has flagged as dual-use: sophisticated reasoning, code generation, autonomous task execution, and the ability to provide technical assistance in domains with weapons or critical infrastructure relevance.
Under the International Emergency Economic Powers Act (IEEPA) and Export Administration Regulations (EAR), the department invoked authority to restrict the transfer of these systems to foreign nationals and foreign-based entities. Because Anthropic's API serves a global customer base, compliance with the order required the company to suspend access for all non-U.S.-based users—and in some cases, suspend access entirely while legal review determines the scope of authorized use.
Anthropic confirmed the shutdown in a brief statement attributed to its policy team, noting that the company is working with the Commerce Department to establish a compliant operating framework, but offered no timeline for restoration of access.
The Models in Question
Fable 5 is Anthropic's most capable reasoning and code generation model, positioned as the successor to the Claude 4 Opus family. Internally benchmarked as exceeding human-expert performance on several software engineering and scientific reasoning tasks, it has been used extensively in enterprise agentic workflows, automated security research, and software development.
Mythos 5 is a specialized model focused on long-context analysis and technical document understanding, with a context window reported to exceed 2 million tokens. It has been deployed in legal, intelligence, and scientific research contexts where processing large corpora of unstructured text is required.
Both models represent capabilities that the government characterizes as qualitatively different from prior generations—specifically pointing to autonomous task completion and the ability to assist in technical processes that historically required specialized human expertise.
Why Export Controls Are Being Applied to AI
The Commerce Department's action follows a broader policy shift toward treating advanced AI models as controlled technologies under export control law—analogous to the way high-performance semiconductors, encryption software, and certain military technologies have historically been regulated.
The logic, as articulated in prior BIS rulemaking, is that sufficiently capable AI systems can lower the barrier to developing weapons of mass destruction, compromise critical infrastructure, or provide adversary nation-states with asymmetric strategic advantages. The agency points to specific capability thresholds—particularly in chemistry, biology, cybersecurity, and autonomous systems—as triggering factors.
Critics argue this framing conflates general-purpose AI capability with specific harmful applications, and that the controls will primarily harm allied researchers and international scientific collaboration without meaningfully restricting adversarial nation-state AI development programs, which operate independently of U.S. commercial AI providers.
Industry Reaction
The reaction from the AI research and technology communities has been swift and largely critical.
Researchers at major universities and international AI safety organizations noted that models with comparable capabilities are already available or under development outside U.S. jurisdiction. The export controls, they argue, do not prevent adversaries from accessing equivalent or superior capabilities but do deprive allied researchers and civil society organizations of tools they use for legitimate purposes.
Several AI policy analysts drew comparisons to the 1990s "Crypto Wars," in which the U.S. government attempted to restrict the export of strong encryption software—a policy that ultimately failed to prevent global adoption of cryptographic tools while creating significant friction for American technology companies and their international users.
Others noted that the speed of the action—and the lack of a clear appeals or licensing pathway communicated to affected customers at the time of shutdown—represents a significant escalation in how the government is choosing to exercise AI-related regulatory authority.
What Comes Next
Anthropic is expected to pursue a licensing arrangement with BIS that would allow access to be restored for vetted, non-adversary international users—similar to frameworks that exist for controlled semiconductors and encryption technology. However, the timeline for such an arrangement is uncertain.
In the interim, users who relied on Fable 5 or Mythos 5 through Anthropic's API are affected. Enterprise customers with existing agreements are working with Anthropic's account teams on fallback options, which in most cases means migrating to Claude Sonnet 4.6 or Claude Haiku 4.5—the models currently unaffected by the export control order.
The decision also raises broader questions about the future of international AI collaboration and the extent to which U.S. regulatory authority will shape the global availability of frontier AI systems. The outcome of Anthropic's licensing negotiations with the Commerce Department is likely to set precedents that affect other frontier AI developers operating in the United States.