Safety and regulation
Who sets the rules, and are they working? · updated 2026-07-30
The EU AI Act is now the world's most complete AI law, phasing in obligations for general-purpose models. The US remains a patchwork of state laws and voluntary frontier-lab commitments. The debate is whether rules this early protect people or just entrench incumbents.
The concern
The concern: voluntary safety frameworks are unenforceable, incidents are reported by the same labs being judged, and capability growth is outpacing evaluation science.
The counter-view
The counter-view: heavy early rules freeze the market for small players (compliance is a fixed cost), the worst harms are already illegal under existing law, and the EU risks regulating an industry it does not host.
Where it stands (July 2026): compliance deadlines are becoming real engineering work; every major lab now publishes a frontier-safety framework; enforcement remains the open question.