A constitution for AI

The values a model holds are real whether or not anyone wrote them down. A constitution, in the loose sense, is just the decision to write them down: to make them explicit, public, and amendable, instead of leaving them implicit and beyond reach.

Does AI need a constitution?

A constitution for AI is a metaphor, and a useful one: the move from values buried in a model's weights to values that are written, visible, and contestable. The point is not a legal document. It is that explicit values can be argued with and amended, and implicit ones can only be discovered by being harmed. Making the values layer for AI legible this way is part of how a system earns trust rather than asking for it, and it lines up with the human-oversight duties arriving as the EU AI Act comes into force.

Why explicit beats implicit

Because you cannot govern what you cannot see. Implicit values are unaccountable by construction: no one signed them, no one can point to them, and no one can change them except by retraining. Written values can be reviewed, challenged, and improved. The act of writing them down is itself a check on power, the same reason real constitutions exist.

The human-oversight duty

A constitution for AI also names who must stay in command. As the law comes into force, meaningful human oversight of high-risk systems shifts from good practice toward obligation, and writing that duty into a system's own charter, not just the regulation, is how it becomes real rather than filed. The constitution is where the values and the oversight meet.