Definition
A proposed US law that would direct the national standards body (NIST) to develop guidance for detecting and preventing 'distillation attacks' — where someone copies a frontier AI model's abilities by training on its outputs. It frames model theft as a national-security and economic-espionage problem.
Why it matters
It signals that protecting model weights and capabilities is becoming policy matter, not just an IT one. Companies that rely on open or frontier models should watch how copying restrictions evolve.