White House orders Anthropic to restrict AI exports: A lesson from history
The U.S. government ordered Anthropic to limit export of its powerful AI models Fable and Mythos over national security concerns. The article draws parallels to past failed attempts to control cyber technology spread, such as PGP encryption and spyware.
Last Friday, the White House issued a directive ordering Anthropic to immediately restrict access to its AI models Fable and Mythos for anyone outside the United States and foreign nationals inside the country, citing unspecified national security concerns. Anthropic complied within about 90 minutes, making both models unavailable to all users for a week.
This incident marks the first real test of whether the U.S. government can use export controls to contain frontier AI. Before the ban, only about 150 vetted companies and government organizations had access to Mythos, which Anthropic marketed as a potential doomsday cyber weapon.
Two events reportedly triggered the ban: first, Anthropic gave a South Korean telecom access to Mythos, and U.S. officials grew alarmed after identifying the company as one suspected of ties to China (the company denied any connection). Second, Amazon CEO Andy Jassy alerted the administration after Amazon researchers found a way around Fable 5’s safeguards, though Anthropic disputed the claim, calling it a narrow, patched issue.
Historically, similar export control efforts have often failed. In the early 1990s, the U.S. government launched a criminal investigation against PGP creator Phil Zimmermann, viewing encryption as a weapon. Zimmermann fought back by publishing the source code as a book, sparking the “Crypto Wars,” and the investigation was eventually dropped, paving the way for modern encryption like Signal and WhatsApp.
In the early 2010s, governments expanded the Wassenaar Arrangement to limit spyware exports. However, the treaty has weaknesses: key countries like Israel are not signatories, and enforcement is left to individual nations. For instance, Italy allowed Hacking Team to export to oppressive regimes. Some spyware makers moved to Saudi Arabia to avoid controls. There were some wins: Germany’s FinFisher shut down in 2022 after an investigation into illegal exports to Turkey.
Currently, the standoff between Anthropic and the Trump administration continues. Possible outcomes include lifting restrictions to keep U.S. AI competitive or requiring government approval for all foreign sales. Given past experience, export controls are unlikely to be the right approach to prevent abuse of dual-use cyber technologies.
