Counter-Agency
Human societies respond to dangerous agency through counter-agency. Laws bind citizens. Neighbors watch neighborhoods. Courts watch police. Inspectors watch factories. Auditors watch firms. Journalists watch governments. Rival powers watch one another, sometimes through treaties and sometimes through the cold discipline of mutually assured destruction. These systems make misalignment more visible, more costly, and harder to hide.
Power cannot become the only witness to itself.
For AI oversight, counter-agency connects technical observation to institutions capable of acting on what is found. A monitor needs a credible reporting channel, an independent assessment of its findings, and a process that can respond. Its performance and the independence of that chain belong in the same safety case.
The requirement also applies to the proposed government. Its scientific judgments need independent challenge. Its jurisdictional thresholds need constraints against self-expansion. Its enforcement decisions require due process and appeal. Its institutions need protection against capture by governments, corporations, ideological movements, and the systems they oversee.
Independence has to appear in appointments, funding, evidence access, removal procedures, and the practical ability to contest decisions. Four offices controlled by the same coalition remain one concentration of power. The inquiry will trace that control and ask who can investigate the institution when its own interests are at stake.