Anthropic restricts AI model exports after White House order

Last Friday, the White House ordered Anthropic to restrict the export of its AI models, Fable and Mythos, to anyone outside the United States and to foreign nationals inside the country, citing unspecified national security concerns. Following the directive, Anthropic complied, hastily pulling access to both models within roughly 90 minutes, by some accounts, of receiving the export control directive—government restrictions on selling technology abroad—issued by the Commerce Department. The restriction has left both models unavailable, marking an escalation in U.S. efforts to regulate dual-use technologies, which have both civilian and military applications.

The export ban reportedly followed two developments. First, Anthropic reportedly gave partner SK Telecom access to Mythos, which previously had around 150 vetted companies and government organizations as users. U.S. officials reportedly suspected SK Telecom had ties to China, though SK Telecom denied any China connection. Second, Amazon CEO Andy Jassy reportedly alerted the administration after Amazon researchers found a way around safeguards in Fable 5. Anthropic disputes the “jailbreak” label for the Fable 5 issue, calling it a narrow, patched issue. Before the ban, Anthropic marketed Mythos as some kind of “Doomsday cyber machine” to help defenders secure software.

This standoff mirrors past attempts to control dual-use software. In the early to mid-1990s, the U.S. government used export controls against Pretty Good Privacy (PGP), characterizing the encryption software as a “dangerous weapon.” PGP creator Phil Zimmermann fought back by publishing PGP’s source code as a book, eventually leading the government to close its investigation. Later, in the early 2010s, the discovery of Western-made spyware in the Middle East led governments to expand the Wassenaar Arrangement, an international treaty limiting the export of dual-use software. However, the treaty has struggled to contain the trade. Italy allowed spyware maker Hacking Team to export tools to “oppressive governments,” and Europe, a bloc of 27 member states, has continually failed to curb spyware exports to “authoritarian regimes.” Some spyware makers, such as the “sanctioned consortium” Intellexa, relocated to countries with lax export controls, while Germany-based FinFisher shut down in 2022 following an investigation into sales to Turkey without an export license.

Currently, the impasse between Anthropic and the Trump administration remains. The outcome could force U.S. AI companies to seek government approval before serving foreign customers, or the administration could lift the restrictions to keep U.S. companies competitive against rivals, including those in China.

Why it matters

This standoff is the first real test of whether the U.S. government can use export controls to contain frontier AI, potentially shaping the rulebook for other AI labs.