From PGP to Mythos: a brief history of export controls that didn't stop anyone
4 hours ago
- #Dual-use Technology
- #National Security
- #AI Export Controls
- The White House ordered Anthropic to restrict exports of AI models Fable and Mythos, citing national security, making them unavailable.
- This is a test of using export controls for frontier AI, similar to past efforts with encryption and spyware, with mixed results.
- Anthropic had tightly controlled Mythos due to its potential risks, but access granted to a South Korean telecom (suspected of China ties) and a reported Fable safeguard bypass triggered the ban.
- Export controls have a history of limited success; for example, the U.S. failed to suppress encryption via the 'Crypto Wars' in the 1990s.
- International efforts like the Wassenaar Arrangement to control dual-use spyware exports have been weak due to non-adherence and lax enforcement by countries like Italy.
- Spyware makers often evade controls by moving to countries with lax regulations, though some have faced consequences, like FinFisher in Germany.
- The outcome of this standoff could impact Anthropic's market access and set rules for other AI labs, with possible outcomes including lifted restrictions or increased compliance burdens.
- Given past software control failures, export controls may not effectively prevent misuse of dual-use AI technologies.