The Evidence Pack: How the US Government Used Export Controls to Halt and Restore Global Access to Anthropic's AI Models

The US Department of Commerce has officially lifted export controls on Anthropic's Claude Fable 5 and Mythos 5 artificial intelligence models, ending an unprecedented three-week global suspension. The resolution marks the conclusion of the first instance in which the US government directly compelled an AI developer to revoke access to deployed commercial models based on national security concerns.[1][2]

The crisis began on June 12, just three days after Anthropic launched the models. The Commerce Department issued an emergency directive under the Export Administration Regulations (EAR), ordering Anthropic to immediately suspend access to the models by any foreign national, whether located inside or outside the United States.

The trigger for the intervention was a security report from Amazon researchers, who discovered a method to bypass Fable 5's safety filters. This "jailbreak" reportedly allowed the model to identify software vulnerabilities and generate code demonstrating how they could be exploited.[1][2]

The directive placed Anthropic in an impossible technical position. Because cloud-based AI providers cannot reliably verify the citizenship of every user in real-time, the company was forced to abruptly disable both Fable 5 and Mythos 5 for all customers worldwide to avoid violating federal law.

The incident highlights a profound shift in how the US government regulates artificial intelligence. Historically, export controls were applied to physical goods like semiconductor chips and manufacturing equipment. A January 2025 rule by the Bureau of Industry and Security (BIS) extended these controls to advanced AI model weights, treating them as dual-use technologies with potential military applications.

The evidence surrounding the severity of the jailbreak remains contested. In its public statements, Anthropic argued that the vulnerabilities identified were minor and could be discovered by other publicly available models without requiring a bypass. The company maintained that its pre-release red-teaming had shown Fable 5's safeguards to be substantially more effective than those of previously deployed models.

"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic stated during the suspension, warning that applying such a standard industry-wide would effectively halt all new frontier model deployments.

However, national security officials viewed the capability through a different lens. Washington has increasingly stepped up oversight of new model releases amid concerns that advanced AI systems could be leveraged by military intelligence in China, Russia, or other adversarial nations to automate cyberattacks or develop novel weapons.[1][3]

The standoff began to thaw on June 26, when the Commerce Department permitted a partial reversal. Anthropic was allowed to release Mythos 5—a model specifically designed to detect cybersecurity vulnerabilities—to more than 100 "trusted" US organizations, including major corporations and government agencies, under a program known as Project Glasswing.[1]

The final resolution arrived late on June 30, when Commerce Secretary Howard Lutnick formally notified Anthropic that the export controls had been withdrawn. "Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI," Lutnick announced.[1][3]

The lifting of the ban was contingent on Anthropic implementing a new safeguard that specifically blocks the jailbreak behavior described in the Amazon report. The company also agreed to proactively detect and address security risks, inform the US government of any malicious activity, and collaborate on protocols for future model releases.[1][2]

Furthermore, Anthropic is now working with Amazon, Microsoft, Google, and other industry partners to develop shared common standards for assessing and patching AI jailbreaks. This collaborative approach suggests a move toward industry-wide self-regulation backed by the threat of government intervention.[1]

Despite the resolution, the Commerce Department explicitly reserved the right to reimpose license requirements if circumstances change or if Anthropic fails to adhere to its commitments. This ongoing conditionality establishes a new paradigm where frontier AI models operate under continuous federal scrutiny even after deployment.[1]

The implications of the Fable 5 suspension extend far beyond Anthropic. The episode has sent a chilling effect through the AI industry, demonstrating that the US government is willing and able to pull the plug on commercial products if it perceives a national security threat.

This new reality is already shaping the deployment strategies of other major players. OpenAI recently delayed the full public launch of its GPT-5.6 model at the government's request, opting instead to limit initial access to a small group of vetted partners while extensive safety testing continues.[3]

The technical challenge of enforcing nationality-based access controls remains unresolved. Cloud security analysts note that as long as AI models are accessed via APIs and web interfaces, distinguishing between domestic users and foreign nationals using VPNs or proxy services will be a persistent vulnerability in the export control framework.

As public access to Fable 5 and Mythos 5 is restored, the AI industry enters a new era of regulatory reality. The rapid iteration cycle that defined the early generative AI boom is now colliding with the slow, cautious machinery of national security, fundamentally altering how the next generation of artificial intelligence will be built and deployed.[2]