What gets regulated, how thresholds are defined, and when intervention occurs
All proposals focus on "frontier" or advanced AI systems, but define the threshold differently. California and New York both use $500M revenue thresholds. The Blackburn bill uses FTC-established thresholds ($500M revenue or 1M+ individuals' data processed). Sanders' moratorium applies to infrastructure rather than models.
The pre-deployment vs. post-deployment question is one of the sharpest divides. OpenAI advocates pre-deployment testing (prevention model). California SB 1047 attempted this but was vetoed; SB 53 retreated to post-deployment transparency. The Blackburn bill creates an evaluation program but the mandatory testing scope is ambiguous. Sanders' moratorium is the most aggressive pre-deployment intervention, stopping the infrastructure itself. The White House framework leans post-deployment (enforcement of existing laws).
Most proposals are horizontal, applying across sectors. Kelly's NDAA provisions are a notable exception, targeting military AI specifically. The Blackburn bill includes sector-specific provisions for children's platforms alongside horizontal obligations.
All proposals in this analysis