AI biosecurity needs more than voluntary safeguards. It needs clear triggers for action before a crisis.