How U.S. Export Controls Forced the Temporary Global Shutdown of Anthropic's Newest AI Models

On June 12, 2026, just three days after Anthropic launched its most advanced artificial intelligence models to date, the systems abruptly vanished from the internet. The sudden disappearance of the Claude Fable 5 and Mythos 5 models was not caused by a technical outage or a server failure. Instead, it was the result of an unprecedented legal directive from the United States government.[1][4][5]

Citing national security concerns, the Commerce Department invoked export control authorities to mandate that Anthropic immediately suspend access to the models for any foreign national, anywhere in the world. This included non-U.S. citizens residing inside the United States, as well as Anthropic's own foreign national employees.[1][4][5][6]

The mechanism of the government's action represents a significant shift in technology regulation. Export controls are traditionally used to prevent the physical transfer of sensitive materials—like uranium, advanced microchips, or aerospace components—across borders. Applying these rules to a cloud-based software model accessed via an application programming interface (API) marks the first known use of export authorities to pull a commercial AI product from public access.[2]

Because Anthropic's platform could not reliably verify the citizenship of its hundreds of millions of users in real-time, the company faced a stark choice. Unable to filter out foreign nationals without risking severe legal penalties, Anthropic opted to disable Fable 5 and Mythos 5 globally for all customers to ensure full compliance with the directive.[1][3][5][6]

The trigger for this sweeping action began with a cybersecurity test. Researchers at Amazon, one of Anthropic's largest investors and its primary cloud infrastructure provider, were evaluating the newly released Fable 5 model. During their testing, they discovered a method to bypass the model's safety guardrails—a technique commonly known in the industry as a "jailbreak."[1][3][5][6]

The specific claim centered on the model's ability to analyze software code. By feeding Fable 5 intentionally flawed code and asking the system to fix the bugs, the researchers found that the model could be coaxed into identifying underlying software vulnerabilities. While defensive cybersecurity teams routinely use such techniques to patch systems, the government viewed the capability as a dual-use risk.[4][6]

If an adversary could use the model to reliably identify zero-day vulnerabilities in critical infrastructure, the AI could effectively serve as an automated cyberattack engine. Recognizing this potential, Amazon CEO Andy Jassy reportedly elevated the findings directly to senior U.S. officials, including Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick.[1][6]

The evidence supporting the severity of the jailbreak remains a point of contention. Anthropic argued that the bypass was narrow and non-universal, claiming it only surfaced minor security flaws that other publicly available AI models could also identify. The company maintained that a single, limited jailbreak did not justify recalling a commercial model deployed to hundreds of millions of users worldwide.[4][5][6]

Despite Anthropic's objections, the government proceeded with the export control directive. The decision underscored a growing anxiety among intelligence agencies regarding the offensive capabilities of next-generation AI. The Five Eyes intelligence alliance recently warned business leaders that frontier models will fundamentally transform both offensive and defensive cybersecurity within months, not years.[2]

The sudden shutdown created immediate ripple effects across the enterprise software ecosystem. Companies that had integrated Fable 5 into their production environments for development and security workflows found their capabilities severed overnight. This highlighted a new vulnerability in the cloud era: reliance on centralized AI models means that access can be revoked instantly by regulatory fiat.[3]

Cybersecurity analysts noted that this dynamic introduces significant concentration and resilience risks. When a vendor is forced to pull a foundational model, the impact cascades across all downstream applications and workflows that depend on it, leaving enterprise defenders scrambling to find alternatives.[3]

After nearly three weeks of intense negotiations, the standoff reached a partial resolution. On July 1, the U.S. government lifted the export controls on Fable 5, allowing Anthropic to restore global access to the model. The company stated that it had reached a series of agreements with the government, though the specific technical or policy concessions remain undisclosed.[2]

However, the more powerful Mythos 5 model remains under tight restrictions. Access to Mythos 5 is currently limited to vetted U.S. organizations through "Project Glasswing," Anthropic's controlled-access program designed specifically for critical infrastructure defenders. Anthropic continues to negotiate for broader domestic and international access to the system.[2]

The uncertainty moving forward revolves around how the U.S. government will standardize this process. Anthropic has publicly called for a statutory framework that is transparent, fair, and grounded in technical facts, warning that applying the current ad-hoc standard across the industry could essentially halt all new model deployments for frontier AI providers.[4][5]

Ultimately, the brief disappearance of Fable 5 serves as a watershed moment in AI governance. It proves that the government is willing and able to use its most powerful regulatory tools to switch off commercial AI products on a few hours' notice, permanently altering the risk calculus for developers and the enterprises that rely on them.[2][6]