U.S. Government Forces Anthropic to Disable Advanced AI Models Under Export Control Order

The U.S. government has crossed a historic threshold in artificial intelligence regulation, using national security powers to force a leading AI developer to pull its most advanced models offline. On the evening of June 12, the Commerce Department issued an unprecedented export control directive to Anthropic, ordering the company to immediately suspend all access to its newly launched Claude Fable 5 and Mythos 5 models by any foreign national. The order treated the cloud-based software as a controlled dual-use technology, effectively classifying the artificial intelligence system as a digital munition subject to strict international embargo. The directive arrived with no public warning, instantly transforming a commercial product launch into a geopolitical flashpoint.[1][2]

Because it is technically impossible to reliably verify the nationality of hundreds of millions of API users in real time, Anthropic was forced to abruptly disable the models for its entire global customer base. The sweeping shutdown occurred just three days after Fable 5 was released to the public, marking the first time the U.S. government has effectively recalled a widely deployed commercial AI system. The sudden outage left enterprise clients scrambling to adjust their software pipelines and ignited a fierce, industry-wide debate over the limits of federal authority in the rapidly accelerating artificial intelligence sector.[1][4]

Fable 5 was specifically designed to make Anthropic's highly capable "Mythos-class" technology safe for general release. The model utilizes a sophisticated layer of separate AI classifiers that detect potential misuse in high-risk areas—such as cybersecurity, biology, and chemistry—before the main model is allowed to respond. If a user's query touches these restricted domains, Fable 5 is programmed to automatically route the request to a less capable, safer model. The fully unrestricted version, Mythos 5, remained accessible only to vetted cyber defenders and infrastructure providers through a private partnership program.[6]

The core of the government's intervention centers on a reported method to bypass these very classifiers. While the Commerce Department, led by Secretary Howard Lutnick, did not publicly detail the vulnerability, administration officials claimed that a trusted partner demonstrated a technique to circumvent Fable 5's guardrails. According to White House AI adviser David Sacks, this "jailbreak" allowed the model to generate unrestricted cybersecurity exploits. Sacks claimed the administration asked Anthropic CEO Dario Amodei to either fix the vulnerability or de-deploy the model, and that Amodei refused, prompting the government's heavy-handed intervention.[2]

Anthropic strongly disputed the government's characterization of the threat, framing the shutdown as a massive overreaction. In a public statement, the company argued that the reported vulnerability was a "narrow, non-universal jailbreak" that only uncovered minor, previously known software bugs. Anthropic noted that other publicly available AI models from rival companies are entirely capable of discovering the exact same vulnerabilities without requiring any complex bypass. The company warned that applying this zero-tolerance standard across the industry would essentially halt all frontier model deployments for every major developer.[1][2][6]

The scope of the Commerce Department's directive was remarkably broad, applying not only to foreign users located abroad but also to foreign nationals legally residing inside the United States. Crucially, the order included Anthropic's own non-citizen employees. Cybersecurity researchers pointed out the operational absurdity of the mandate: the very software engineers and researchers who built the Fable 5 model were suddenly legally prohibited from looking at the code they had written, severely hampering the company's ability to actually study and patch the alleged vulnerability.[1][5][8]

Legal and policy analysts note that the government likely relied on the Export Controls Reform Act of 2018, which authorizes the Commerce Department to restrict the transfer of dual-use technologies that could aid adversaries. While the U.S. has previously used these powers to block the physical export of advanced semiconductor chips and manufacturing equipment to geopolitical rivals like China, applying them to an intangible, cloud-hosted software service accessed via an API represents a massive and legally untested expansion of regulatory authority. By treating an API call as an export, the administration has effectively claimed jurisdiction over the real-time data flows of the global internet, setting a precedent that could ensnare countless other cloud services.[7]

The shutdown arrives against the backdrop of a rapidly deteriorating relationship between Anthropic and the Trump administration. Earlier this year, Anthropic formally refused to allow the U.S. military to use its AI models for fully autonomous weapons systems and domestic surveillance programs. In retaliation for that refusal, the Pentagon placed Anthropic on a federal supply chain blacklist. Several industry observers and tech analysts view the sudden export control directive not just as a security measure, but as a severe escalation of this ongoing political feud.[3][4]

For the broader cybersecurity community, the incident evokes strong historical parallels to the "Crypto Wars" of the 1990s. During that era, the U.S. government attempted to classify strong encryption software as a military munition to prevent its export overseas. Those efforts ultimately failed because the underlying mathematical knowledge proliferated globally regardless of American law. Security researchers warn that treating AI models as munitions may face the exact same fate, noting that mathematics and open-source code do not stop at customs checkpoints.[5][8]

The immediate commercial impact on Anthropic is severe, striking the company at a highly sensitive moment in its corporate trajectory. Anthropic had recently filed a confidential IPO prospectus with the Securities and Exchange Commission, disclosing a revenue run rate of $47 billion and a staggering valuation of nearly $1 trillion. Fable 5, priced at $10 per million input tokens, was marketed as the company's flagship commercial product and a key driver of future revenue. While Anthropic's older models, such as Claude Opus and Sonnet, remain online and unaffected, the forced withdrawal of its newest tier creates immense financial uncertainty and threatens to derail one of the most anticipated public offerings in tech history.[2][6]

Moving forward, the precedent set by the Fable 5 shutdown fundamentally alters the risk calculus for the global technology industry. If any application depending on a U.S.-based large language model can be switched off overnight by a unilateral export control directive, the reliability of American AI infrastructure is permanently compromised. Analysts predict that foreign governments and multinational corporations will rapidly accelerate their shift toward sovereign, self-hosted, and open-source AI alternatives to ensure their operational continuity is never again subject to the stroke of a Commerce Department pen.[7][8]