The AI Ad-Hoc Prior Restraint Era Begins

The White House has directly intervened in Anthropic's deployment plans for its frontier AI model Mythos, blocking the expansion of Project Glasswing that would have given more companies access. This unprecedented move signals the beginning of an ad-hoc prior restraint era in American AI policy, where the government informally decides which models can be made available to which parties. The decision comes amid concerns about cyberattacks and the need for defenders to have advanced AI tools, but the lack of formal regulatory framework raises serious concerns about arbitrariness and potential abuse.

Critics, including Neil Chilson and Dean Ball, argue that this informal approach is worse than a structured regulatory regime, as it favors insiders and creates uncertainty. The White House is now considering an executive order to establish an AI working group to examine procedures for vetting models before release. While some form of prior restraint may be necessary for frontier AI, the ad-hoc implementation risks stifling innovation and creating confusion, especially as labs like Anthropic face pressure from both domestic security concerns and European Union demands for access.