A US export-control order forced Anthropic to disable its newest AI models worldwide. The move followed reports of a jailbreak exposing advanced cybersecurity features. Industry impact could be significant.
On June 12, 2026, just three days after Anthropic unveiled its most advanced AI models, the US Commerce Department issued a sweeping order: Fable 5 and Mythos 5 had to be taken offline for all users worldwide. Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei, instructing the company to suspend access to both models for any foreign national, including those working within the United States. Within hours, both models were disabled for every customer globally.
This action marks one of the most forceful uses of US export-control authority against a commercially available AI system. Fable 5, Anthropic’s flagship public release, and Mythos 5, a less restricted version available only to select organizations through Project Glasswing, were both removed from service. Other Claude models—Opus, Sonnet, and Haiku—remain accessible.
The government’s decision was triggered by reports of a jailbreak that could bypass Fable 5’s safety guardrails, potentially exposing the powerful cybersecurity capabilities embedded in the Mythos architecture. Officials feared that if these protections could be circumvented, the models could be misused as unrestricted cyber tools.
Axios reported that the administration had previously tried to prevent Anthropic from launching the models, but after those efforts failed, the export-control order followed. In a statement, Anthropic confirmed the directive, citing national security concerns and explaining that the company had to abruptly disable both models to comply.
White House AI adviser David Sacks posted on X that a trusted partner testing Fable 5 had discovered a jailbreak and alerted both Anthropic and the government. According to Sacks, the administration asked Amodei to either fix the vulnerability or withdraw the model, but Amodei declined. Sacks described the government’s response as reluctant and said the next steps were up to Anthropic.
Anthropic disputes this version of events, arguing that the reported jailbreak demonstrated capabilities already present in other public models, such as OpenAI’s GPT-5.5, and that cybersecurity professionals routinely use such tools. The company stated it was not given details about the alleged vulnerability and denied refusing to address issues. As of publication, the two sides have not reconciled their accounts.
While the identity of the “trusted partner” has not been officially confirmed, reporting by Semafor and the Wall Street Journal points to Amazon. Sources say Amazon researchers used Claude Fable 5 to extract information relevant to cyberattacks and reported their findings to US officials, prompting the export-control action. Amazon, a major Anthropic investor and cloud provider, declined to comment on specifics but acknowledged that governments often seek its input on security risks.
Another factor reportedly influencing the White House’s decision was concern that a China-linked group may have accessed Mythos, raising fears of reverse engineering or model distillation by foreign adversaries.
The shutdown comes amid ongoing tensions between Anthropic and the Trump administration. The Department of Defense previously labeled Anthropic a supply chain risk after failed negotiations, leading to a lawsuit from the company that remains unresolved. Defense Secretary Pete Hegseth recently referenced the dispute on X, defending the Pentagon’s decision to bar Anthropic from its facilities.
Some have speculated that these prior conflicts influenced the export-control order, but Sacks rejected that idea, insisting the action was unrelated to earlier disputes.
Timing has also drawn attention. One day after launching Fable 5, Amodei published a policy essay advocating for US government authority to block or reverse AI models that fail independent safety tests. Two days later, the government exercised that very power against Anthropic.
The order arrives at a sensitive commercial moment for Anthropic, which recently filed a confidential IPO prospectus with the SEC, disclosing a $47 billion revenue run rate and a $965 billion valuation. The precedent of a government disabling a widely used AI product mid-deployment over a disputed security assessment is likely to reverberate across the industry.
In his essay, Amodei compared the pace of AI development and regulation to the mismatched speeds of Hobbits and Treebeard in The Lord of the Rings. This week, he wrote, “Treebeard moved fast.”
Anthropic, founded in 2021 by former OpenAI researchers, has rapidly become a major player in the AI sector. The company’s Claude model family is used by millions worldwide, and Anthropic counts Amazon and Google among its largest investors. As of June 2026, Anthropic’s confidential SEC filings revealed a $47 billion annual revenue run rate and a $965 billion valuation, underscoring its influence and the high stakes of regulatory intervention. (Source: Forbes)
