Consent must remain effective
Democratic authority depends on institutions that people can understand, contest, and change. AI could help people participate by improving access to information and public services. It could also make consequential decisions harder to challenge if responsibility disappears into automated processes.
A government may formally remain in charge while relying on systems its officials cannot independently evaluate or replace. Citizens may retain voting rights while losing effective influence over how important decisions are made.
A decision with no reachable decision-maker
Suppose an automated service denies a family access to an essential benefit. The appeal office repeats the system's explanation, lacks authority to override it, and cannot reconstruct the reasoning. The problem is not simply an incorrect answer. The family has no effective path to correction.
At a larger scale, powerful actors could use AI for surveillance or tailored manipulation. Autonomous systems could present a different problem if institutions become unable to redirect them. These routes can overlap, but obedience to a government and obedience to democratic public control are not equivalent.
Keep power contestable
Public decisions need identifiable responsibility, accessible appeals, independent investigation, and a real possibility of reversal. People also need protected space to organize without pervasive monitoring. Essential institutions should retain the expertise to evaluate providers and switch away from them.
Proposals to restrict dangerous AI must meet the same test. Verification can target relevant computing resources and deployments without treating universal surveillance as an acceptable default. No institutional design guarantees freedom forever. The useful question is whether a proposed system gives ordinary people more effective ways to challenge power, or quietly removes the ones they have.