On June 12, 2026, the US Commerce Department ordered Anthropic to immediately suspend access to its most advanced AI models, Claude Fable 5 and Claude Mythos 5, for all foreign nationals anywhere in the world. The directive, issued by Commerce Secretary Howard Lutnick under the Export Administration Regulations, cited national security concerns over a reported jailbreak technique that could bypass the model’s safety guardrails. Unable to enforce the restriction selectively by nationality, Anthropic disabled both models globally for all customers, marking the first time the US government has used export controls to shut down a commercially deployed frontier AI model.
What Are Claude Fable 5 and Mythos 5?
Anthropic is an American AI safety and research company founded in January 2021 by former OpenAI executives, including siblings Dario Amodei (CEO) and Daniela Amodei (President). Headquartered in San Francisco, California, the company is best known for its Claude family of large language models, built using a technique called constitutional AI, which trains models to adhere to a set of guiding principles.
On June 9, 2026, Anthropic released two new models. Claude Fable 5 was the public-facing model available to all customers through the Claude platform and API. It was built on the same underlying technology as Mythos 5 but with safety guardrails restricting its most powerful capabilities in cybersecurity and biotechnology. Claude Mythos 5 was the full-capability version, restricted from the start to a small group of vetted partners through a programme called Project Glasswing.
Project Glasswing, launched in April 2026, is a collaborative cybersecurity initiative bringing together Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Under this programme, Mythos Preview (and later Mythos 5) was made available to cybersecurity defenders to scan critical software infrastructure for vulnerabilities. Within weeks, partners had found more than 10,000 high- or critical-severity security flaws across critical systems.
Fable 5 was described by Anthropic as state-of-the-art on nearly all benchmarks, with particularly strong performance in software engineering, scientific research, and autonomous task execution. It was the company’s first Mythos-class model made available to the general public.
What Triggered the US Government Action?
The trigger was a reported technique for bypassing Fable 5’s safety guardrails, an attack known as jailbreaking. The concern was that if the guardrails could be defeated, a consumer-facing AI product would effectively become an unrestricted cyber tool capable of identifying and exploiting software vulnerabilities at scale.
According to Anthropic, the government believed it had become aware of a method to bypass Fable 5’s safeguards. However, the company strongly disputed the severity of the finding. In its public statement, Anthropic said it had reviewed the demonstration and found that the technique identified “a small number of previously known, minor vulnerabilities” and that “other publicly available models are able to discover them as well without requiring a bypass.” The company specifically pointed to OpenAI’s GPT-5.5 as another model capable of the same task.
Anthropic stated that it received only “verbal evidence of a potential narrow, non-universal jailbreak” which essentially consisted of asking the model to read a specific codebase and fix any software flaws. The company said the government’s letter did not provide specific details of its national security concern. White House adviser David Sacks offered a different account, claiming that a trusted partner testing Fable 5 came forward with a jailbreak, that Anthropic was asked to fix it, and that the company refused. “The ball is in Anthropic’s court,” Sacks wrote.
Adding to the pressure, Amazon CEO Andy Jassy reportedly alerted Treasury Secretary Scott Bessent and other officials that Amazon researchers had used Fable 5 to obtain information that could be used in cyberattacks.
What Did the Export Control Directive Say?
The directive was issued by Commerce Secretary Howard Lutnick in a letter to Anthropic CEO Dario Amodei. It was received by Anthropic at 5:21 PM ET on June 12, 2026. The letter ordered Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national employees of Anthropic itself. A license from the Bureau of Industry and Security (BIS) would now be required for the export, reexport, or domestic transfer of either model.
The legal basis for the order is the Export Control Reform Act of 2018 (ECRA) codified in 50 U.S.C. 4801 et seq. ECRA provides the legal framework for US dual-use export controls, administered by BIS through the Export Administration Regulations (EAR). The EAR governs the export of goods, software, and technology that have both commercial and military applications.
Anthropic concluded that it could not reliably verify the nationality of every user in real time. As a result, the company said it had to “abruptly disable” both models for all customers worldwide to ensure full compliance. Access to other Anthropic models, including Claude Opus 4.8, was unaffected.
The directive represented the first known use of US export control authorities to regulate a specific frontier AI model on national security grounds. It also applied the concept of a deemed export to AI: traditionally, deemed export rules cover the transfer of controlled technical information to foreign nationals within the US. The Anthropic order extended this principle to operational access to a deployed AI model.
Why This Sets a Precedent
This action has significant implications beyond the two models involved. Previously, US export controls on AI focused on hardware (advanced semiconductors) and, since January 2025, on model weights under the AI Diffusion Rule. The Anthropic directive marks a shift from controlling physical or digital artifacts to controlling operational access to AI capabilities delivered as a service.
For the AI industry, the move raises fundamental questions. Anthropic argued that if the government’s standard was applied across the industry, it “would essentially halt all new model deployments for all frontier model providers.” The company noted that the same vulnerability existed in other publicly available models that were not subject to similar restrictions.
The action also intensified an already fraught relationship between Anthropic and the US government. Earlier in 2026, Anthropic had refused to permit the US military to use its models for domestic surveillance and fully autonomous weapons. In response, the Pentagon designated Anthropic as a supply chain risk, and the company sued the Defense Department. While some tensions had reportedly eased, this export control directive showed the core conflict remained unresolved.
On the global stage, the shutdown triggered alarm across Europe, Canada, and other regions. Companies, research institutions, and public services that had integrated Fable 5 into their operations suddenly lost access, not because of any action by their own governments, but because of a directive from Washington. The episode underscored a stark reality: access to frontier AI capability is now subject not just to market forces but to the jurisdiction of the country that hosts the technology.
Critics also pointed to the opaque nature of the process. Anthropic said it received no written explanation of the specific threat, and the government has not publicly disclosed the directive or its full justification. This lack of transparency, some legal experts argue, sets a concerning precedent for due process in AI regulation.
The Way Forward
Anthropic has stated that it is complying with the directive and working to restore access as soon as possible. The most likely path to reinstatement involves addressing the government’s concerns about the jailbreak vulnerability and obtaining clearance from BIS through individually validated licenses.
David Sacks framed the off-ramp clearly: fix the jailbreak, lift the control. Anthropic has reportedly already met with White House officials to find a resolution. The company maintains that the vulnerability was narrow and that applying this standard to the entire industry would be unworkable.
Separately, Project Glasswing is expected to expand. Anthropic has indicated its intention to broaden the trusted access programme to cover more organizations in more countries, with explicit export control carve-outs for cybersecurity defence work. This approach would allow Mythos-class capabilities to reach vetted defenders while addressing government concerns about uncontrolled proliferation.
The episode also raises broader questions about how AI governance should work. Anthropic itself has previously called for a statutory process that is “transparent, fair, clear, and grounded in technical facts” for blocking unsafe AI deployments. The company argued that the current action did not adhere to those principles. The outcome of this dispute will likely shape the future of AI regulation, both in the United States and globally, as other countries grapple with how to oversee increasingly capable AI systems.
Key Takeaways
- The US Commerce Department issued an export control directive on June 12, 2026, forcing Anthropic to suspend global access to Claude Fable 5 and Claude Mythos 5 just three days after their public launch.
- The directive was issued by Commerce Secretary Howard Lutnick under the Export Administration Regulations (EAR), with the legal authority derived from the Export Control Reform Act of 2018 (ECRA).
- Anthropic was founded in January 2021 by former OpenAI executives, including Dario Amodei (CEO), and is headquartered in San Francisco, California.
- Project Glasswing, launched in April 2026, is a cybersecurity initiative under which Mythos 5 was made available to vetted partners for defensive vulnerability scanning.
- This is the first instance of the US government using export control authorities to regulate a commercially deployed frontier AI model, marking a significant shift in AI governance.
- The Bureau of Industry and Security (BIS) in the Department of Commerce administers dual-use export controls under the EAR, and now requires individually validated licenses for access to the restricted models.
