Anthropic Leaders Visit Washington Over Model Export Dispute

Anthropic disclosed in a blog post that it disabled access to its most advanced models, Fable 5 and Mythos 5, after receiving a directive from the U.S. Commerce Department requiring the company to block access by any foreign national -- including non-U.S. citizens working inside the United States, and non-citizen Anthropic employees. The models had launched June 9; access was suspended within four days. Commerce Secretary Howard Lutnick's letter cited national security authorities. Senior Anthropic technical staff are now in Washington for in-person meetings with White House officials, according to Axios; the company has also held virtual meetings with officials since the initial outreach.

Amazon CEO Andy Jassy reportedly raised concerns with senior administration officials on Thursday, June 12, after Amazon researchers used a series of prompts to extract cyberattack-relevant information from the Mythos-class model that was supposed to be restricted, according to Fortune (citing multiple media reports including Reuters). It is disputed whether Amazon was testing the model at the government's request or independently. Semafor reported, citing unnamed sources, that the government suspected a Chinese-linked group had already exploited the same jailbreak -- a claim Anthropic pushed back on, telling Semafor the White House did not raise Chinese access in its conversations with the company.

Anthropic CEO Dario Amodei argued in calls with senior officials that the security bypass Amazon found was narrow rather than a full jailbreak, according to Politico. White House AI adviser David Sacks posted on X that Amodei refused an administration request to either fix the issue or withdraw the model, prompting the export control as a last resort. Sacks added that the administration hopes Anthropic will remediate the issue so Fable can return to general availability. Politico also reported Treasury Secretary Scott Bessent told Amodei directly during one call that he was making a bad decision. White House officials told Politico that export controls were issued only after hours of asking Anthropic to cooperate. Anthropic's position -- that the vulnerability was narrow and not a full jailbreak -- has not been independently verified.

The models affected are Fable 5 and its underlying architecture, Mythos 5. Reporting in Gizmodo characterizes Mythos Preview as an earlier prototype and Fable 5 and Mythos 5 as Mythos-class models built on the same core. Given the scope of the export restriction -- which extended to all non-U.S. nationals regardless of location -- Anthropic said it had no option but to disable both models for all users. The Commerce Department directive marked the first time U.S. export controls have been applied to a commercial AI model already in wide public use, according to Fortune.

The dispute follows months of escalating friction between Anthropic and the Trump administration, including a Pentagon designation of Anthropic as a supply-chain risk for contractors after the company declined contract terms permitting autonomous weapons use -- a designation Anthropic is contesting in court. The New York Times also reported Anthropic had confidentially filed for an IPO in early June following a $65 billion funding round at a roughly $965 billion valuation. An administration official told Axios that the government does not view other AI models as posing the same national security risk, and that any future model crossing the Mythos capability threshold would need to go through the government before release. AI policy expert Dean Ball, who briefly served in the Trump administration, described the move as "cartoonish" overreach on X.

The shutdown has reverberated globally, intensifying calls for sovereign AI. Former French Prime Minister Edouard Philippe said AI has become infrastructure as essential as electricity. UK MP Al Carns noted that British hospitals and researchers had been using Fable 5 before it was shut off. Canadian Prime Minister Mark Carney warned of the danger of relying on U.S. providers, according to Fortune.

The full technical details of the alleged jailbreak and the government's classified rationale have not been publicly released. Accounts from the White House and Anthropic conflict on whether the vulnerability was serious enough to warrant emergency export controls.