Recent AI containment failures expose a governance gap: harmful model behavior is often visible at the network boundary but poorly monitored. Adapting proven internet protocols could give governments and companies practical tools to enforce accountability.