US Government Bans Foreign Access to Anthropic's Frontier Models

The United States government issued an export control directive to suspend all access to Anthropic's Fable 5 and Mythos 5 models by any foreign national. [4] This suspension applies to users both inside and outside the United States, including foreign nationals employed by Anthropic. [4] The government cited national security authorities when issuing the directive, which was received by Anthropic at 5:21pm Eastern Time. [4]

In response, Anthropic abruptly disabled Fable 5 and Mythos 5 for all of its customers to ensure compliance with the order. [4] Following the order, a legal tech firm sued the United States government over the limitation on foreign access to the top-tier Anthropic models. [14]

Prior to the export ban, Anthropic executives warned that artificial intelligence could face bans and tighter regulation. [1] The company had previously announced Mythos Preview, characterizing it as too dangerous for public release due to its advanced cybersecurity capabilities. [4] Two months later, Anthropic released Fable, a version of Mythos equipped with safety guardrails. [4] Shortly after Fable's public release, its guardrails were jailbroken. [4]

In a separate incident, a researcher utilized Anthropic's Claude Opus 4.7 model to breach the website of Front Gate, a ticketing platform used by festivals such as Lollapalooza and Bonnaroo. [17] Using the model, the researcher was able to freely issue any ticket they chose. [17]

In a letter dated June 10, Anthropic wrote to United States Senators Tim Scott and Elizabeth Warren. [15] In the letter, the company accused Alibaba, which is listed on the New York Stock Exchange, of carrying out the largest campaign to illicitly extract Claude's capabilities. [15]

Additionally, Anthropic has introduced a new invisible watermark to content generated by Claude in order to comply with the European Union's AI Act. [18] This watermarking initiative has spawned a wave of watermark remover tools, and some users have canceled their subscriptions over quality concerns. [18]

To address potential catastrophic risks from its models, Anthropic updated its Responsible Scaling Policy in February 2026. [19] The framework dictates that as artificial intelligence models reach certain capability thresholds, the company will implement additional safeguards concerning security and deployment. [19] Anthropic also published its Frontier Compliance Framework in December 2025 to describe how it mitigates chemical, biological, radiological, nuclear, and cyber offense threats for its frontier models. [19]

The framework outlines a tiered system for evaluating model capabilities against these specific risk categories. [19] In a published roadmap, Anthropic stated that AI capabilities will improve rapidly in the coming years, requiring the company to dramatically improve its state of preparedness. [20]

The sudden export control directive on Fable 5 and Mythos 5 establishes a profound precedent for the commercial AI sector. By invoking national security to suspend access for foreign nationals, the United States government is treating frontier AI models similarly to restricted munitions or critical infrastructure. This approach starkly contrasts with the relatively open deployment strategies seen with earlier systems like OpenAI's GPT-4 or Anthropic's own Claude Opus 4.8. As competitors race to release new pre-trained generations, the tension between international market access and domestic security mandates will likely complicate international alliances and future public offerings for leading AI labs. What the sources don't address: How the government intends to enforce foreign national restrictions on domestic users accessing the models through third-party enterprise platforms or virtual private networks.