Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9, 2026. Three days later, both were gone. Not because of a technical failure, but because of a letter.
Commerce Secretary Howard Lutnick sent that letter to Anthropic CEO Dario Amodei on the evening of June 12. It ordered Anthropic to suspend all access to both models for any foreign national, whether inside or outside the United States. That scope included Anthropic's own non-American employees. Because the company cannot verify the citizenship of millions of users in real time, it did the only thing it could: it shut both models off for everyone.
The story has two triggers, not one. The widely reported trigger was a security claim by Amazon researchers, who said they found a way to get Fable 5 to provide information useful for cyberattacks. Amazon CEO Andy Jassy told Treasury Secretary Scott Bessent. The White House gave Anthropic a 90-minute ultimatum and imposed controls when Anthropic pushed back, arguing the technique was narrow and worked equally well on rival models including OpenAI's GPT-5.5. Seventy-six cybersecurity experts signed an open letter calling the export order unjustified, saying the action had taken the best tools away from defenders without meaningfully reducing risk.
The less-reported trigger came first. The White House had already grown alarmed that SK Telecom, South Korea's largest wireless carrier and a $100 million Anthropic investor since 2023, had been granted access to Mythos through a controlled partner program called Project Glasswing. Officials cited SK Telecom's historical ties to China Unicom, a state-owned Chinese carrier: the two companies ran a joint venture from 2004, and SK Telecom invested $1 billion in China Unicom bonds in 2006, later converted into roughly a 6.6 percent equity stake. By 2021, the first Trump administration had restricted US investment in China Unicom on national security grounds. When the White House told Anthropic to cut off SK Telecom, Anthropic complied immediately. The government's trust was already gone before the Amazon call.
The result is a company the US government simultaneously considers too dangerous for its own use and too dangerous for anyone outside America to use. Earlier this year, the Pentagon sought unrestricted access to Claude for mass surveillance and autonomous weapons. Anthropic refused. Defense Secretary Pete Hegseth designated the company a supply chain security risk, a label normally reserved for foreign adversary firms like Huawei. Anthropic sued, and that case is still active.
For every other AI lab, the lesson is plain. Companies now believe they need to give the White House early access to models before launch. The Trump administration signed an executive order in late May creating a voluntary pre-release review system, but after what happened to Anthropic, that regime is effectively mandatory, just undeclared. One former White House technology official put it directly: the White House said it was not a licensing regime, and it now clearly is.
Amazon's role is uncomfortable. The company has invested heavily in Anthropic and its own researchers triggered the crackdown on its own investee. Amazon said it is not uncommon for governments to seek its counsel on security risks, but offered no further explanation of why Jassy made the call he made.
The IPO timing adds another layer. Anthropic filed confidential public listing paperwork on June 1, eleven days before the shutdown. The company had just closed a funding round at a reported $965 billion valuation, with business subscriptions surpassing OpenAI's for the first time in May. Interestingly, data from business spending platform Ramp suggests the drama may not be hurting Anthropic's brand: its best month on record for enterprise adoption was the month the Pentagon labeled it a supply chain risk. Prediction market traders placed roughly 58 percent odds on Fable 5 returning before July 1. But public-market investors weigh government exposure differently than business customers do, and the shutdown landed directly on the IPO story.
June 12, 2026 is the first time the US Commerce Department used the 2018 Export Control Reform Act against a live, deployed AI model. The precedent is set. Export controls, once understood as a hardware issue about chips and server components, now apply to software running in a browser.
Outside the US, the reaction skipped the jailbreak debate. At the G7 summit in France, French President Emmanuel Macron told Trump and assembled AI executives directly that if the US can turn off the switch from one day to the next, nobody will buy American AI. India's Prime Minister Modi raised the same concern. G7 leaders discussed a trusted-partners scheme that would let allied nations retain access to advanced American models, but the details remain entirely unresolved. In India, entrepreneurs and policy experts compared the moment to Russia's loss of access to the SWIFT financial messaging system after its Ukraine invasion: a reminder that even embedded infrastructure can be switched off by geopolitical decision.
For any business outside the US that had built workflows around Fable 5 or Mythos 5, the practical lesson is already clear. Tools delivered through the cloud by a US company are subject to US law, not yours. Any provider headquartered in the United States can receive an order like this one. If your business runs critical work through any of them, you now know what that looks like: a letter arrives on a Friday afternoon, and by evening the tool is gone.