GenAI Governance Needs Non-User Participation

Participation is not endorsement

This is not an argument that everyone should use GenAI. A person or organization may have good reasons not to adopt a particular tool, or to reject a particular use.

It is an argument for something narrower: people who will be affected by an AI decision should have a way to influence the rules around it, whether or not they use the system themselves.

That position has practical support. The NIST AI Risk Management Framework says risk management should reflect diverse, multidisciplinary perspectives and may include AI actors outside the organization. It also describes governance as a continuous activity across the system lifecycle. (NIST AI RMF Core)

The important distinction is simple. Personal non-use can be an ethical choice. It is not, by itself, a governance process.

Who is missing from the room?

An AI project team is likely to hear first from people who chose to build or buy the system. They are necessary voices, but they cannot be the only ones. A decision may also affect people whose work is evaluated by it, people whose data is used by it, customers who receive its output, and staff who must explain its mistakes.

Those people might be heavy users, occasional users, or non-users. The point is not to assign every person a vote on every technical detail. It is to make sure the people carrying the consequences can identify risks that a project team may not see.

Skeptics are useful here. They may ask whether the problem needs this system at all, who bears the cost of errors, what information should never be used, and what decision must remain reviewable by a person. Those are governance questions, not signs of obstruction.

A practical participation loop

For a consequential use, participation can be modest and concrete:

1. Name the decision. State what the system will recommend, automate, or prioritize, and what will remain a human decision.
2. Map the affected people. Include users, non-users, frontline staff, people represented in the data, and people who will handle appeals or corrections.
3. Invite relevant challenge. Bring in domain, privacy, security, legal, accessibility, worker, and affected-community perspectives where they fit the decision.
4. Record the response. Keep a short record of the concerns raised, the limits adopted, the dissent that remains, and who owns the next decision.
5. Reopen the question. Monitor what happens after deployment. A complaint pattern, an unexpected error, or a changed use case can be a reason to revisit the decision.

This is not a substitute for legal obligations, technical testing, or a formal risk program. It is a way to stop governance from being defined only by the people most invested in moving quickly.

Staying involved while staying critical

Participation does not require liking the technology, building products, or defending a vendor. It can mean asking for transparency, recommending a limit, declining a use case, documenting a concern, or insisting on a review path for people affected by an outcome.

It also does not require accepting every invitation to participate. Some forums will be performative, and some decisions will not be open to meaningful influence. In those cases, a person may choose public criticism, organizing, regulatory channels, or non-participation. The claim here is only that, when a real decision forum is open, informed criticism can improve the decision more than silence can.

An invitation with boundaries

GenAI is current, visible, and moving quickly. That is enough reason to take its governance seriously; it is not a reason to declare every use inevitable or every critic an enemy of progress.

The useful question is not, “Are you for AI or against it?” It is, “Who gets to help set the limits, and what happens when those limits fail?”

People who do not use a system may be among the best people to ask that question. They should not have to become enthusiasts to be heard.