A partnership with South Korea's largest mobile carrier and a separate vulnerability report from Amazon drove the Trump administration to revoke global access to Anthropic's Claude Mythos model and its public spinoff, Fable 5.
When Anthropic released its most powerful AI model to the public on June 9, 2026, the company believed a heavily guarded version called Fable 5 could safely put frontier AI in people's hands. Five weeks later, the model was gone from every country on Earth, pulled because a single commercial partnership with South Korea's largest mobile carrier had set off a cascade through Amazon's security researchers, the White House, and the Commerce Department's export enforcers.
The chain, as reconstructed by WIRED, runs through four distinct links. Anthropic granted SK Telecom access to Claude Mythos, the company's most capable model. Amazon's security researchers told the White House they had found vulnerabilities in Fable 5, the publicly released version, that could let users bypass guardrails, the safety constraints meant to keep the model from producing dangerous outputs, and reach Mythos's cybercapabilities, the offensive hacking functions built into the frontier system. The Trump administration sent Anthropic a letter demanding the company cut off all foreign nationals. Anthropic, rather than attempt nationality-gated access, pulled the model entirely.
Each link looked manageable on its own. Together, per a person close to the administration cited in WIRED's reporting, they produced a judgment that Anthropic could not be trusted to safeguard its most advanced AI.
The SK Telecom deal was the original spark. SK Telecom is South Korea's largest mobile carrier, a sensitive conduit for frontier AI in a US ally with deep economic ties to China. US officials alleged that SK Telecom had ties to Chinese interests. That allegation is the government's framing, and SK Telecom's on-record position on the claim is not detailed in the public reporting available for this article. What is in the record is that an allied carrier now held direct access to Mythos, and that access alone was enough to put Anthropic on the administration's radar.
Then Amazon's researchers delivered their own separate report. The Amazon team told the White House they had identified vulnerabilities in Fable 5 that could be exploited to circumvent guardrails and reach Mythos's underlying cybercapabilities. Anthropic and outside cybersecurity experts argue, on the record, that the circumvention risks Amazon flagged are not unique to Claude. The administration did not treat that pushback as dispositive.
The White House letter that followed did not, per WIRED's account, explicitly cite SK Telecom or China. Its stated pressure mechanism was the nationality question: who, exactly, would be allowed to use the most powerful version of Mythos once Fable 5 had demonstrated the possibility of sliding past its constraints? Anthropic's response was to remove the model from the market worldwide rather than build a system that admitted US persons and excluded everyone else.
What makes the cascade unusual is its trigger. It was not a hack, a safety failure, or a Chinese espionage operation. It was a commercial revenue deal, a vulnerability disclosure from a rival cloud and AI business, and a regulatory apparatus that had never before been aimed at a single AI model. The administration chose export controls, the same legal tool used to keep advanced semiconductors out of certain hands, as the pressure mechanism. That choice, more than any individual vulnerability finding, is what converted a partnership dispute into a global model shutdown.
Several pieces of the chain remain unresolved. WIRED's reporting relies on anonymous US government and company-adjacent sources, and the SK Telecom and China allegation is the government's framing rather than an established fact. Amazon's circumvention findings are contested by Anthropic and outside researchers. The Commerce Department's order, dated to a Friday relative to WIRED's publication but not pinned to a specific calendar date in the public reporting, will set the precedent for whether future AI models can be sold abroad at all. And SK Telecom's on-record response to the China tie allegation is not detailed in the available record.
What to watch next is whether the export-control order survives the inevitable court challenge, and whether Anthropic's decision to pull rather than gate becomes the template for other frontier labs facing the same question: build a system that decides who is allowed to use the most powerful AI in the world, or refuse to ship it at all.