On the evening of June 12, 2026, the artificial intelligence world was rocked by an unprecedented move from the United States government. Anthropic, the California-based AI company behind the widely-discussed models Fable 5 and Mythos 5, announced it had received a sweeping export control directive from the U.S. Department of Commerce. The order, delivered at 5:21 PM Eastern Time, required Anthropic to immediately suspend access to both models for all foreign nationals—regardless of whether they were inside or outside U.S. borders—including even the company’s own foreign employees.
Within hours, Anthropic complied, disabling Fable 5 and Mythos 5 for every user worldwide. According to Axios, this was an extraordinary step: never before had a leading AI company been forced by federal authorities to pull already-launched commercial models offline on such short notice. The impact was instant and global. Developers and enterprises from Seoul to San Francisco suddenly found themselves locked out of what had been, just days earlier, the most advanced public AI models on the market.
So what sparked this dramatic government intervention? The answer, it turns out, is both technical and political—and, as Anthropic would argue, deeply contentious. The U.S. Commerce Secretary Howard Lutnick had sent a letter to Anthropic CEO Dario Amodei, citing national security concerns. According to government sources reported by Axios and Yonhap News, a third-party company had claimed to have breached Mythos 5’s security barriers using so-called “jailbreak” techniques—methods for tricking AI models into bypassing their built-in safety controls. This, officials said, raised fears that the models could be exploited for malicious purposes, from cyberattacks to aiding in the creation of biological or chemical weapons.
But Anthropic was quick to push back. In a series of public statements, the company insisted that the reported jailbreak was extremely limited in scope—nothing more than a technical “workaround” already known to exist in other leading models, such as OpenAI’s GPT-5.5. “The government has only provided verbal evidence of a narrow and non-generalized jailbreak, such as instructing the model to read a specific codebase and fix software bugs,” Anthropic stated. “The same functionality is used daily by defenders on other public models, including GPT-5.5.”
Anthropic further explained that, before launching Fable 5 and Mythos 5 on June 9, it had conducted thousands of hours of rigorous red-team safety testing in collaboration with the U.S. government, the UK AI Safety Institute, and multiple independent organizations. These tests, the company claims, failed to uncover any widespread vulnerabilities or “generalized jailbreaks.” Fable 5, in particular, was designed with layers of robust safety features: it automatically rerouted sensitive queries related to cybersecurity, biology, or chemistry to a more restricted higher-tier model, Opus 4.8. Mythos 5, meanwhile, was a less restricted version made available only to select participants in Project Glasswing, a global cybersecurity research initiative that included partners like Samsung, SK Hynix, and the Korea Internet & Security Agency.
The government’s directive, however, left little room for negotiation. It required that Anthropic block access for all foreign nationals, inside and outside the U.S., and warned of financial and civil penalties for violations. The company’s only option, given the impossibility of surgically excluding every foreign user, was to suspend service for everyone. This included South Korean organizations and developers, who suddenly found themselves cut off from Fable 5—prompting South Korea’s Ministry of Science and ICT to reach out to Anthropic for clarification, according to News1.
The reaction from Anthropic was a mix of frustration, apology, and warning. “We believe this directive is based on a misunderstanding,” the company said, expressing regret for the confusion and inconvenience caused to its customers. Anthropic promised to share additional information and work toward restoring access within 24 hours, though it admitted the timeline and the full impact on users remained uncertain. “Recalling a commercial model already distributed to hundreds of millions of users on the basis of a very limited jailbreak possibility is unreasonable,” the company argued. “If this restrictive standard were applied across the industry, the release of all new frontier models would effectively stop.”
Anthropic’s stance was clear: perfect prevention of jailbreaks is impossible. “All security measures used in the industry are vulnerable to certain jailbreak methods in specific situations,” the company explained. “We implemented a defense-in-depth strategy, making successful jailbreaks extremely costly and maintaining rigorous monitoring so that any attack could be detected and blocked quickly.” The company also pointed out that the types of jailbreaks cited by the government were not unique to Anthropic’s models—they were, in fact, common across the field and regularly used by cybersecurity professionals for legitimate purposes.
Despite the technical debate, the government’s position remained firm. Yonhap News reported that the Department of Commerce decided to act after a company claimed to have breached Mythos’ security. Prior attempts by the government to persuade Anthropic to delay launching the models had failed, so the export control directive was issued as a last resort. Officials suggested that the suspension would likely last several weeks, until U.S. national security protocols could be strengthened.
For Anthropic, this wasn’t the first time it had clashed with the U.S. government over its AI models. Earlier in the year, the company had refused Department of Defense requests to open up its Claude models for use in fully autonomous weapons and mass surveillance, with CEO Dario Amodei declaring, “I simply cannot accept this on principle.” The latest export control drama, however, marked a new level of regulatory confrontation—one that could have far-reaching implications for the global AI industry.
In the days following the suspension, the conversation extended far beyond Silicon Valley. In South Korea, where Anthropic had just opened a new office and appointed a local representative, the Ministry of Science and ICT began communicating with the company to understand the full scope of the export controls. Project Glasswing participants, including major Korean tech firms, were left in limbo, unable to access the Mythos 5 models they had been using for cybersecurity research.
Meanwhile, the AI community and industry observers are watching closely. If the government’s approach becomes the new norm, it could set a precedent that stifles innovation and slows the release of advanced AI models worldwide. Anthropic’s warning is hard to ignore: “Applying such a restrictive standard industry-wide would halt the deployment of all new frontier models.”
As the debate over AI safety, national security, and regulatory oversight intensifies, one thing is certain—what happened to Fable 5 and Mythos 5 this June will be studied for years to come as a pivotal moment in the evolving relationship between government and artificial intelligence.