US Government Orders Anthropic to Suspend Advanced AI Models Over Security Concerns

In an unprecedented escalation of technological regulation, the United States government has forced artificial intelligence company Anthropic to abruptly disable its most advanced models globally. The shutdown, triggered by an emergency export control directive, marks a historic shift in how Washington polices the frontier of artificial intelligence.[2][7]

At 5:21 p.m. Eastern Time on Friday, June 12, Commerce Secretary Howard Lutnick issued a letter to Anthropic CEO Dario Amodei. The directive ordered the immediate suspension of access to Anthropic's newly released Fable 5 and Mythos 5 models for any foreign national, regardless of whether they reside inside or outside the United States.[1][3]

Because the order applied to all non-US citizens—including Anthropic's own foreign-born employees and researchers—the company concluded that selective compliance was technologically impossible. Unable to instantly verify the citizenship of millions of users and API developers, Anthropic took the drastic step of pulling Fable 5 and Mythos 5 offline for everyone, including American citizens.[1][5]

The shutdown comes just three days after Anthropic launched the models to widespread industry acclaim. Fable 5 was designed as the company's most capable general-release model, featuring a massive one-million-token context window and advanced reasoning capabilities. Mythos 5, built on the same underlying architecture, was a specialized version with certain safety classifiers removed, available only to a vetted group of cyber defenders and critical infrastructure operators.[1][4]

The core of the government's concern revolves around a cybersecurity vulnerability. According to Anthropic, the Commerce Department acted on a tip from a third-party company claiming to have found a "jailbreak"—a method of bypassing the model's safety guardrails.[3][4]

The alleged jailbreak involves prompting the model to read a specific codebase and identify software flaws. While the government viewed this as a severe national security risk, Anthropic has publicly disputed the severity of the finding. The company characterized the vulnerability as a "narrow, non-universal jailbreak" that relies on simple, previously known techniques.[1][2]

Anthropic argues that the capability to identify these specific software flaws is not unique to Fable 5. The company stated that it reviewed the demonstration underlying the government's order and found that other publicly available models, including OpenAI's GPT-5.5, can execute the exact same tasks without even requiring a bypass.[1][3]

This directive represents a watershed moment in US technology policy. For years, Washington's strategy for containing foreign adversaries' AI capabilities focused almost entirely on hardware—specifically, restricting the export of advanced Nvidia microchips and semiconductor manufacturing equipment.[2]

By targeting cloud-based access to a deployed model, the Commerce Department has effectively classified the API outputs of a frontier AI system as a controlled export. This closes a perceived loophole where foreign actors could simply rent computing power and intelligence from American servers, but it also introduces massive friction into the global software ecosystem.[2][7]

The government's aggressive posture reflects growing anxiety within the defense establishment about the dual-use nature of advanced AI. The Pentagon's chief information officer, Kirsten Davies, publicly supported the move, stating that national security must take precedence over "revenue cycles, clickbait, and pre-IPO valuation."[2]

The shutdown also arrives against a backdrop of existing friction between Anthropic and the US defense apparatus. Earlier this year, Anthropic refused to allow the US military to use its models for domestic surveillance or fully autonomous weapons systems. In response, the Department of Defense labeled the company a "supply chain risk," a designation Anthropic is currently fighting in court.[2][4]

The abrupt global outage has immediately validated the arguments of "sovereign AI" advocates. Companies outside the United States are pointing to the shutdown as proof that relying on American-hosted, closed-source models is an existential business risk.[6]

Isaacus, an Australian foundational AI research company, noted that the directive affects developers in allied nations like the UK, Canada, and Australia just as severely as those in adversarial states. "In effect, any application depending on US-based LLMs is subject to being shut down at any moment as a result of an export control directive," the company stated, arguing for a pivot toward air-gapped, self-hosted models.[6]

Moving forward, the directive raises immense logistical questions for the AI industry. Dean Ball, a former White House official, noted that enforcing such an order long-term would require companies to implement stringent "Know Your Customer" (KYC) protocols. "This means you should expect to have to prove your citizenship to use Anthropic models," Ball observed.[2]

Anthropic has warned that the government's rationale, if applied consistently, would paralyze the entire sector. Because no AI model is perfectly immune to jailbreaks, the company argued that using a narrow bypass as grounds for a recall "would essentially halt all new model deployments for all frontier model providers."[1][3]

For now, Fable 5 and Mythos 5 remain inaccessible globally. Anthropic has stated it is working to resolve what it calls a "misunderstanding" with the Commerce Department to restore access. However, the precedent has been set: the US government has demonstrated both the willingness and the legal mechanism to pull the plug on the world's most advanced software systems with zero warning.[1][5]