A post in the series on independent AI auditing, picking up from Who Decides What Gets Measured.
Dominique Charlie Oquendo
Founder and Principal Consultant, Intentional Data. November 2026.
Claude, an AI model made by Anthropic, helped with research and editing. The argument and the writing are my own. Ask for the repo to see how Claude was used!
Introduction
A Stanford course hosted by Stanford's McCoy Family Center for Ethics in Society 1 brought together AI practitioners to discuss and contribute to thinking around ethics, technology and public policy. Over the past seven weeks, I have had the privilege of learning in this space.
Rob Reich, the Stanford professor who opened the course, began with Ursula K. Le Guin's story "The Ones Who Walk Away from Omelas." A whole city is happy, joyous and carefree because one child is kept out of sight, tortured and abused. Some residents of Omelas who learn this truth walk away in disgust, leaving the child where it is 2. As we discussed whether the residents who walked away from Omelas were cowards or heroes, Reich reminded us of the important autonomy we have as human philosophers. We have the privileged ability to think for ourselves and reject the premise of any argument.
Since then, I have begun attempting to digest my learnings around AI accountability and AI policy through writing. My first two posts focused on an accessible analysis of policies around auditor requirements and audit criteria. With the closeout of the ethics course comes another layer to consider in my own writing and understanding: when and how community voice is considered in AI policy.
Key learnings
Three learnings from the course have stayed with me.
An audit is only as good as who it answers to. In the course of the 7 weeks, we were introduced to a number of articles, research papers, books, and ways of thinking. One was a research paper by Raji, Xu, Honigsberg and Ho. The paper found that when a company picks and pays its own auditor, the result "may be functionally indistinguishable from first-party audit." This means an audit done by a paid outside source is no different from an audit done by the company itself 3. That finding matched what I had already seen in the law. Illinois is the only state that requires an audit; however, the only thing the auditor checks against is the company's own internally created safety framework 4. Imagine if you were the one creating a test that you had to pass.
Design choices are policy choices. Our October 21 session made the point that product design and launch timelines are policy choices. What a model learns from, what it is built to predict and when it ships all decide who is protected and who is hurt, and those choices are made before any audit begins. Limiting the personal data a model trains on is one example, or keeping personal data out of the training pipeline from the start. Preventing a harm in the design costs less than removing it after launch.
People affected by an AI tool need a say before it is deployed. In our October 28 session, arguably the most hopeful of sessions, we talked about involving workers in procurement before a tool is deployed. Three examples show how differently people can be given a say. Pennsylvania's Generative AI Labor and Management Collaboration Group was formed in March 2025 with members of SEIU Local 668 and Department of Human Services management, and it exists to "give employees a greater voice in shaping ongoing AI adoption" 5. National Nurses United published a bill of rights for nurses and patients that lists collective advocacy for workers and their patients among its principles 6. The Writers Guild contract says a company cannot require a writer to use AI and must tell the writer when material it hands over was generated by AI 7. Each gives people a different kind of footing: a seat, a right or a notice.
Since many affected workers, especially early-career workers, are not in a union, people need more than one way into these decisions. That is why I count unions, worker groups and community organizations as the same kind of voice. These groups are where people can shape what trains AI, what gets examined and when, and how we define AI safety.
The idea of community voice in AI safety is why the open door in California's registry excites me. AB 1405 directs the state's Government Operations Agency to open an AI Auditor Registry by January 1, 2029 8. The law defines an AI auditor as a person, partnership or corporation, and I understand this as allowing a group to register. It names no required certification, and it asks that the people assigned to an audit collectively possess the knowledge, skills and abilities the audit needs. I read that as room for a coalition of community groups and technical auditors to register together and bring what a community knows about a system's real effects into that expertise.
Application
The goals I have around AI policy and coproduced technology have manifested themselves in several ways over the past few months. As previously mentioned, I have begun writing accessible policy analysis, I am pursuing a certification in algorithmic auditing, and, most importantly, I am facilitating a free community training around AI literacy and governance. The training gives community organizations access to shape the skills and tools they will use in their daily operations with AI technology. The tools that come out of it are co-created with them and planned for public release.
The Stanford course happened alongside all of that, and it gave me something I did not expect. I had spent a long time feeling like I was doing this work alone, with no backing, no fellowship and no PhD program behind me. Seeing my peers, who are real-world practitioners, left me replenished and rejuvenated, and it made me want to keep dreaming and pushing toward what I care about.
My first goal for 2027 is to organize community members to take part in the City of Austin's AI safety and labor policies. The Texas legislative session starts in January and is an open door for collaboration. The City has already done meaningful work here, including three Community in the Loop sessions, facilitated by my partner organization Measure. The City calls this an early step in a longer process 9.
My second goal is to support people who want to get into the work of keeping AI safe and accountable, through certification access, fiscal sponsorship as needed and as funds allow, and support while they learn. I also want to understand what it would take for the people a system decides about to be part of decision making before the list of what gets examined is set, with a say in what counts as a finding.
Conclusion
An audit designed without the people it affects will check a system against a definition of safety that does not include us. The fictional city of Omelas that kicks off our course gives its residents two options: to stay or to walk away. Reich's reminder tells me I do not have to accept that those are the only two ways of being. A third option exists, and it requires building a way for the child to have a say. My work with Austin residents and with people entering this field is my first step toward that.
If you have been inside an audit, a participatory evaluation or a community review board, or your group already has a say in how AI is chosen, write to me at dominique.charlie@intentionaldata.org and tell me where.
Endnotes
- Stanford Ethics in Society. 2026. Ethics, Technology and Public Policy for Practitioners. ethicsinsociety.stanford.edu
- Le Guin, U. K. 1973. The Ones Who Walk Away from Omelas. New Dimensions 3. ursulakleguin.com
- Raji, I. D., Xu, P., Honigsberg, C., and Ho, D. E. 2022. Outsider Oversight: Designing a Third Party Audit Ecosystem for AI Governance. AIES 2022. arxiv.org/abs/2206.04737
- Illinois General Assembly. Senate Bill 315. ilga.gov
- Commonwealth of Pennsylvania. March 21, 2025. Shapiro Administration Leads the Way in Ethical Use of AI. pa.gov
- National Nurses United. 2024. Nurses and Patients' Bill of Rights: Guiding Principles for A.I. Justice in Nursing and Health Care. nationalnursesunited.org
- Writers Guild of America. 2023 Minimum Basic Agreement summary and Artificial Intelligence page. wga.org
- California. September 9, 2026. Assembly Bill 1405, Chapter 178 of 2026. leginfo.legislature.ca.gov
- City of Austin. September 2026. Advancing Austin's Open Government Commitments. austintexas.gov