Washington’s attempt to box in AI mirrors failed tech export bans

The directive forced Anthropic to shutter access to its models within 90 minutes of notification, leaving roughly 150 vetted organizations in the dark. The clampdown followed reports that a South Korean telecom partner allegedly maintained ties to China, coupled with internal warnings from Amazon CEO Andy Jassy regarding potential model jailbreaks. Anthropic maintains the latter was a minor, patched vulnerability rather than a systemic security failure.

History suggests such mandates face significant headwinds. In the 1990s, the U.S. government attempted to classify encryption software as a weapon, launching a criminal probe against PGP creator Phil Zimmermann. The effort collapsed when Zimmermann published his source code as a book, effectively cementing encryption as a protected form of speech. Similarly, the Wassenaar Arrangement, designed to curb the global spread of commercial spyware, has been undermined by inconsistent enforcement and the ease with which firms like Intellexa relocate to jurisdictions with lax oversight. While Germany’s prosecution of FinFisher marked a rare victory against illicit spyware, the broader reality remains that software rarely stays contained by national borders. As the Trump administration weighs its next move, the standoff highlights a fundamental tension: if Washington enforces these restrictions, it risks hobbling American AI companies; if it retreats, it concedes that global competitors will inevitably bridge the capability gap.