Anthropic Export Ban Mirrors Failed PGP Crypto Wars
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic pulled its AI models Fable and Mythos after the White House ordered export restrictions on both, citing unspecified national security concerns; the models have been unavailable to anyone for roughly a week.
- The Commerce Department issued the directive after two reported triggers: a South Korean telecom (widely reported as SK Telecom, which denies China ties) gained access to Mythos, and Amazon CEO Andy Jassy alerted the administration that Amazon researchers found a way around Fable 5's safeguards — a finding Anthropic disputes as a narrow, already-patched issue.
- Anthropic had marketed Mythos as a "Doomsday cyber machine" capable of wreaking havoc on the internet if released widely, and before the ban limited access to roughly 150 vetted companies and government organizations.
- The U.S. government tried to stop PGP encryption in the early 1990s by opening a criminal investigation against creator Phil Zimmermann for violating arms export controls; Zimmermann published the source code as a printed book, the investigation was closed, and end-to-end encryption later became the basis for Signal and WhatsApp.
- The Wassenaar Arrangement, expanded in the early 2010s to cover dual-use surveillance software, has failed to stop spyware proliferation because non-adhering countries like Israel aren't bound by it and member states like Italy have self-policed loosely — Hacking Team exported tools to oppressive governments despite the rules.
- FinFisher shut down in 2022 after a multi-year German investigation into alleged illegal spyware sales to Turkey; sanctioned consortium Intellexa and other spyware makers simply relocated operations to countries with lax export controls.
- The article flags two likely outcomes: the administration lifts the restriction to keep U.S. AI firms competitive — a tacit acknowledgment that Chinese labs will match the capabilities — or AI companies face a permanent compliance burden of needing government approval before serving foreign customers.
Why it matters: This is the first real test of whether the U.S. can use export controls to contain frontier AI the way it has tried — and largely failed — to contain encryption and spyware. With PGP and the Wassenaar Arrangement as precedent, the controls risk either saddling U.S. AI labs with a compliance burden on foreign sales or tacitly conceding China will match their capabilities regardless.
