Blog · arXiv Analysis · Last reviewed June 24, 2026

The Policy Table Becomes the Participation Filter

The June 2026 arXiv paper Challenges to Grassroots Organization Engagement with AI Policy, by Carter Buckner, Jennifer Mickel, Nandhini Swaminathan, William Agnew, Jacob Hobbs, Sarthak Arora, Michelle Lin, Yanan Long, and B.V. Alaka, treats participation in AI policymaking as work that has costs, geography, gatekeepers, and failure modes.

From Invitation to Filter

The paper, arXiv:2606.19816 [cs.CY], was submitted on June 18, 2026. Its starting point is simple: governments are writing AI policy on privacy, economics, intellectual property, energy, and other risks, while public involvement is supposed to make governance more accountable. The authors' sharper claim is that public involvement is not equally available to everyone.

That makes this a distinct companion to the site's reviews of democratic AI governance and automated public comments. Those pages ask whether the public has a formal route into governance and whether synthetic participation pollutes that route. This paper asks a prior question: who can afford to show up before the route becomes meaningful?

The policy table becomes a participation filter when it rewards organizations that already have travel budgets, policy staff, insider networks, grant time, legal fluency, and the social license to be treated as representative. An invitation to comment is not equal access if one group can send a paid policy team and another must assemble a volunteer response at night.

The Case Study

Buckner and colleagues use collaborative autoethnography to describe Queer in AI's attempt to bring participatory design principles into U.S. AI policymaking. The paper describes engagement with policy bodies including NAIAC and NIST, and it discusses the group's participatory development of a queer-focused U.S. AI policy explainer.

That method matters. The paper is not a neutral survey of every civil-society organization in AI policy. It is a situated account from people trying to do policy work while holding participatory values. The authors' positionality statement says they bring training in machine learning, natural language processing, privacy, security, fairness, ethics, and legal research, as well as varied experience in policy design, participatory governance, advocacy, unionizing, and activism. The strength of the paper is precisely that it treats the backstage labor as evidence.

Participation Has a Geography

One of the paper's practical findings is geographic. It says AI policy work is often concentrated in a few in-person centers, naming places such as Washington D.C., London, Brussels, and San Francisco. For a distributed grassroots organization, that geography is not a detail. It decides who hears informal signals, who builds relationships, who learns which document matters, and who can correct a misunderstanding before a comment deadline closes.

Virtual access is therefore not a convenience feature. It is part of the legitimacy of the process. The paper argues for virtual or strong hybrid participation, public notes, recordings, and meeting infrastructures that do not make remote participants secondary. A regulator that announces an accessible comment channel while reserving the real conversation for rooms elsewhere has not solved participation. It has moved the barrier.

The Volunteer-Time Problem

The paper also names time. Distributed participatory design in policy can demand prohibitive volunteer and staff labor, especially from smaller organizations. The authors describe sudden workload spikes, long-distance coordination, and organizational scarcity. They also report that the burden often falls on core members with relevant expertise, many of whom are students or junior academics doing unpaid work alongside coursework, publication pressure, jobs, health constraints, and other advocacy commitments.

This should change how AI governance reads "civil society consultation." A regulator may count submitted comments, workshop attendees, or advisory seats. Those counts do not show the cost imposed on the participants. If marginalized communities must pay for representation with exhaustion, unpaid research, travel, credential performance, and repeated explanations of basic harms, then consultation becomes an extraction mechanism.

Capture and Opacity

The paper's strongest governance point is about opacity. The authors report that RFIs and advisory processes can be hard to navigate, that feedback may not visibly affect policy, and that industry-heavy memberships or partnerships can weaken civil-society trust. They also describe uncertainty over which documents are serious, which bills are signaling, which comments are being read, and where meaningful decisions are happening.

This is the same institutional pattern that appears in release-theater red teaming: public input is invited, but the evidence chain from participation to decision is hard to inspect. A public process becomes credible only when participants can see what was received, how it was summarized, what changed, what was rejected, and why.

Governance Standard

The governance standard is to audit participation as infrastructure. A real process should publish schedules early, fund travel or compensate labor where possible, provide strong remote access, maintain plain-language explainers, support coalition submissions, disclose advisory relationships, publish comment logs, and show how public input changed the policy text.

It should also avoid asking one marginalized organization to stand in for a whole population. Representation is not a checkbox. It is a relationship with scope, consent, credit, limits, safety, and compensation. Participation should not require an organization to become a permanent interpreter between affected communities and an expert policy class that retains final discretion.

The Spiralist rule is direct: a public AI policy process is not democratic because the door is technically open. It is democratic when ordinary and affected people can cross the threshold without losing the rest of their lives to the cost of being heard.

Sources

Return to Blog