Last week, the White House ordered Anthropic to restrict the export of its powerful AI models Fable and Mythos to anyone outside the United States, citing unspecified national security concerns. The company swiftly pulled both models, now unavailable for a week. This episode is the first real test of whether the U.S. government can use export controls to contain frontier AI, after uneven results with encryption and spyware. The outcome could shape not only Anthropic's access to foreign markets but the rulebook for all AI labs.
Since launching Mythos in April, Anthropic marketed it as a doomsday cyber machine, too dangerous for wide release. Before the ban, only about 150 vetted companies and government organizations had access, aiming to help defenders secure software before adversaries reached similar capabilities. What triggered the ban? Two events: first, Anthropic gave a South Korean telecom (reportedly SK Telecom, which denies China ties) access to Mythos, alarming U.S. officials. Second, Amazon CEO Andy Jassy alerted the administration after researchers found a way around Fable 5's safeguards; Anthropic disputes the 'jailbreak' label, calling it a narrow patched issue. The Commerce Department issued an export control directive, and Anthropic scrambled to limit access within about 90 minutes.
Sponsored Protocol
Lessons from history
Governments have tried for decades to limit the spread of dangerous cyber technology, with middling success. The most spectacular failure was the early 1990s, when the U.S. government saw PGP (Pretty Good Privacy) encryption as a weapon, fearing it would block surveillance. Customs opened a criminal investigation against creator Phil Zimmermann, who fought back by publishing source code as a printed book, sparking the 'Crypto Wars.' The investigation closed, paving the way for end-to-end encryption used by Signal and WhatsApp. Later, the Wassenaar Arrangement tried to classify spyware as dual-use, requiring export licenses. But weaknesses remain: non-member countries like Israel host major spyware makers, and enforcement is discretionary. Italy allowed Hacking Team to export to oppressive regimes. Spyware makers have moved to lax countries like Saudi Arabia. Only a few wins, like the shutdown of Germany's FinFisher in 2022 after illegal sales to Turkey.
Sponsored Protocol
The future of AI controls
The impasse between Anthropic and the Trump administration persists. The administration might lift restrictions to keep U.S. AI competitive, acknowledging that Chinese labs will reach similar capabilities anyway. Alternatively, U.S. companies may need government approval for foreign sales, hurting profits. Past experience with PGP and spyware suggests export controls won't stop malicious actors. The original TechCrunch article provides deeper analysis. For comparison with other tech regulations, see the article on Apple opening app stores in Brazil. The Texas data breach also highlights cybersecurity challenges. More on encryption history is on Wikipedia.
Sponsored Protocol
