Frontier AI governance & accountability

This is the governance stack for very large models: regulators want to know which models exist (registry), what happened when a frontier model causes harm (incident reporting and attribution), and to hold someone accountable (failure to prevent offence). Combined with fail-safe shutdown, the goal is to make frontier AI traceable, stoppable, and answerable to a human.