Whose values should AI hold?
Every AI model encodes values, so the only honest question is whose, and who decides. Human values in AI is the problem of choosing those values, legitimising them, and keeping a human in the loop on the decisions that express them. It is not a soft topic: when a model declines one request, approves another, or flags a person, it is applying a judgment about what matters more. Sanctity makes that judgment explicit and shared, which is the values-layer half of human judgment infrastructure.
Why does AI have values at all?
Because building a model is a series of choices: what data to include and leave out, what to optimise for, what to forbid. Each is a small value judgment, and the finished system is their sum. Calling a model neutral does not remove its point of view, it only hides it. The honest move is to name the opinion and ask whose it is, the argument in every model is an opinion.
Whose values, and who decides
On questions of value, the people affected should weigh equally; on questions of fact, expertise should weigh more. That distinction decides a great deal. There is no expert who is simply right about what a system should care about, so values call for a public square, not a panel. The OECD AI Principles and the Council of Europe's work on AI both push toward this kind of legitimacy. The fuller case: whose values should AI hold.
Keeping a human in the loop
Choosing values is not enough; a human has to stay in command of the decisions that apply them. That is where human values in AI meets human oversight, and where the EU AI Act's Article 14 makes oversight a legal duty for high-risk systems. Values and oversight are two halves of the same trust, the two layers of human judgment infrastructure.
Where Sanctity fits
Sanctity is the values layer for AI: a public square where everyone, equal, helps decide what AI should hold, paired with an expertise layer where an agent's hardest calls reach an accountable human. Human judgment infrastructure, built so the values are chosen in the open and the oversight is real.
Read next
- The values layer for AI
- EU AI Act Article 14: human oversight, explained
- Trustworthy AI: what makes AI trustworthy
- AI governance: the frameworks and where oversight fits