Comment by Thomas Holland

The people who first see a dangerous training run are usually inside the lab, under NDAs, watching a launch calendar. Whistleblower protection is how those people become a warning system instead of a silence system. [...] Frontier AI researchers have a vantage point no regulator or inspector currently matches, and they are frequently bound by contracts and non-disparagement clauses that make speaking out costly or career-ending. Protecting their ability to warn is one of the most tractable and underrated levers in ASI governance. [...] Protecting the insiders who can warn us in the meantime is something legislatures can do now, and it is among the highest-leverage safety measures available before a treaty exists.