The White House has ordered Anthropic to restrict exports of its powerful AI models, Fable and Mythos, following a series of events that alarmed U.S. officials. This marks the first real test for export controls on frontier AI technologies, which have had a patchy track record in containing other digital dangers.

The decision comes after Anthropic provided access to SK Telecom, suspected by U.S. officials to be linked to China, and after Amazon researchers reportedly found ways around Fable 5’s safety measures. The company swiftly complied with the Commerce Department directive within 90 minutes of being notified.

Historically, governments have struggled to effectively regulate dangerous cyber technologies through export controls. In the early 1990s, the U.S. government tried to stop the distribution of PGP encryption software, leading to what is now known as the 'Crypto Wars,' where Phil Zimmermann published the code and won a key legal battle.

The Wassenaar Arrangement, an international treaty aimed at limiting the export of dual-use technologies like spyware, also faces challenges. Israel, home to active spyware makers, does not adhere to it, while lax enforcement by other countries continues to allow spyware exports to authoritarian regimes.

While some progress has been made, such as Germany’s FinFisher shutting down after an investigation into selling unlicensed spyware, questions remain about whether export controls can keep up with the rapidly evolving landscape of frontier AI technologies.