Who decides?
Substantive / decision power. Who may choose, authorize, reject, or modify the final public action?
Today · N0
A proposed governance architecture for the AI age
Better public decisions, with AI under human authority.
NousPolis is designing a constitutional architecture through which public institutions could use AI as analytical capacity without making AI a source of political authority. Citizens would help define public problems. Multiple AI systems and human reviewers would examine evidence, assumptions, alternatives, and uncertainty. Legitimate human institutions would retain the authority and responsibility to decide.
AI can research, compare, forecast, and challenge. It cannot authorize public policy.
Current reality: NousPolis remains an N0 architecture and research project. An experimental executable policy-process pilot now exists, but there is no deployed governance institution and no public-policy authority.
Illustrative example
Start with a public question
Should a city allow taller housing around railway stations?
Residents might raise affordability, displacement, accessibility, transport, environmental, infrastructure, and neighborhood concerns. Different analytical approaches could examine those claims without collapsing them into one model answer. Evidence, assumptions, conflicts, and uncertainty would remain inspectable. The legitimate public institution—not the AI system—would still authorize any policy.
The deeper constitutional problem
Keeping the final vote human is not enough.
Authority can be exercised before the final decision. NousPolis therefore treats three forms of power as separate design problems.
Who controls what gets considered?
Procedural / process power. Who frames the question, routes work, selects analysis, aggregates findings, stops the process, and controls the record?
Who decides which safeguards apply?
Definitional / rule-setting power. Who assigns classifications such as impact, materiality, credibility, or catastrophic risk that trigger stronger safeguards?
The implication: keeping the last vote human is insufficient if the question, the evidence, or the safeguards were controlled elsewhere.
Proposed
Evidence → analysis → deliberation → authorization
Different responsibilities must remain genuinely different.
Technical strength is not democratic consent, and a recommendation is not permission to act.
Public and affected communities
Raise issues, contribute testimony and local knowledge, challenge framing and omissions, inspect reasons, and use lawful participation or appeal routes.
AI systems, researchers, and reviewers
Trace evidence, build competing interpretations, analyze institutions and coalitions, forecast consequences, expose uncertainty, and adversarially challenge conclusions.
Legitimate human institutions
Define lawful authority, protect rights, resolve value conflicts through legitimate procedure, authorize action, and remain accountable for implementation and results.
Today · N0
What exists now
The architecture has begun to become machine-testable—without crossing the N1 gate.
Architecture v0.1 defines the institutional rules and limits. On 17 August 2026, the main repository merged the hardened experimental Policy Deliberation and Learning Standard (NP-PDLS-001) executable pilot v0.1. Its reference validator and adversarial suite test a structured policy process, including classification, evidence, independent analysis, adversarial review, participation, safeguards, human decision records, monitoring, and reopening.
What this does not mean: the PDLS pilot does not establish N1 maturity, external independence, evidentiary truth, legal authority, policy correctness, or authorization for real-world policy action. It remains an N0 test instrument.
The latest published N1 gate snapshot remains dated 13 August 2026: 9 of 11 grouped requirements passed, the overall result was FAIL, and N1 remained blocked.
Stewardship today
Founder-led, founder-funded, and disclosed.
NousPolis is currently founded and solely funded by Deon Van der Westhuizen. Founder control is a bootstrap condition, not the intended institutional end state.
Why it matters
AI can make analysis abundant. Legitimate institutions remain scarce.
More analytical capacity can help institutions examine more evidence, compare more alternatives, preserve disagreement, and learn from outcomes. It can also concentrate hidden influence in whoever controls models, questions, classifications, evidence admission, or stopping rules. NousPolis starts with those institutional limits rather than with product features.
Inspect and challenge the work