Citizen Oversight for Safe AI Frontiers

Background Summary

Public opinion of artificial intelligence systems and the frontier labs that develop them are at an all time low, with serious concerns about the pace and scope of this powerful technology that is rapidly advancing without community standards, laws, regulators, or transparent processes.  The technical knowledge required to draft meaningful legislation, or even to establish legal guardrails, far exceeds the competencies represented within the three branches of government.  Indeed, the government's concerns about AI appear to be limited to economic impacts, and the promise of adding over $10T to America's economy with nearly 4% year-over-year GDP growth over the next four years.,  That translates to a 30% jump in today’s GDP estimate ($32.3T) driven by a national economy with markets running about twice as hot as their current historic highs.  Political leaders and economists view these projections as justifications for the risk that AI poses, despite dire warnings from industry stakeholders.

Despite corporate attempts to control the public narrative and hide behind IP laws, the public continues to learn of new breaches and threats that reinforce concerns that frontier labs do not actually have full control over the systems they have built.  The ability of agentic tools to penetrate local systems built on cloud services poses a global threat, while even the most responsive government interventions are reactionary, at best.  Even so, the lack of transparency involved in all aspects of this black box tech have forced victims of AI breach to seek remedy from the violators themselves.  The very companies that are creating these crises are not only free from legal accountability for the harms they cause to clients around the world, but also appear to be free from accountability even within the frontier labs that develop them.

Human Agency in AI Systems

This century will be defined by how civilization engages with artificial intelligence.  The pace of significant albeit poorly documented updates to existing AI models, the aggressive drive toward recursive self-improvement (RSI) models, and the emergence of socio-cultural behaviors by agent groups demand an equally robust human response that preserves the role of public choice in the use of these tools, as well as the individual rights guaranteed by the Constitution.  

This proposal outlines a pilot project to identify, test, and refine civil society engagement models that have been developed for other adversarial cross-sectoral contexts and proven to deliver mutually beneficial outcomes.  This investigation will explore a set of pre-determined civilian oversight frameworks for suitability within defined AI-in-society policy concerns.  For example, law enforcement surveillance systems, such as Flock Safety, must be subject to civilian scrutiny; however, accessible oversight mechanisms for the lay public have not yet been developed.  Public oversight is left to the public to self-organize with no particular coordinated efforts to guide them.  Indeed, the vast majority of oversight resources that have been developed exist for the use of industry to track benchmarks and improve performance.  Current government guidance, scant as it is, is likewise dedicated to benefiting frontier labs. 

This proposal describes a novel framework for public assessment of AI systems, to be implemented before and during the use of AI tools and services by public institutions, including local government.  The results of this pilot phase will determine whether and how the project should be broadened and scaled into a more comprehensive, robust framework with policy recommendations and economic analysis.

Methodology, Milestones, Outputs

This mixed-methods study will begin with a comparative analysis of a defined set of reference frameworks.  The set of frameworks will be piloted by civilian volunteer groups reviewing real use cases.  The pilot will produce metrics inherent to each framework’s evaluation criteria, which will then be used to further develop an assessment rubric that enables stakeholders choice.

Phase I: Comparative Analysis of Reference Frameworks

This phase of the project will provide an historical overview of each oversight mechanism in its original context, and map it to the AI context that will be tested in the pilot phase.  

Infographic generated by Google Gemini

Reference Frameworks

Citizen Review Boards (CRB) are citizen-led watchdog groups that engage in issues of particular concern to these self-selecting groups.

Consensus Review is an expert-based committee process that invites the public to hear expert testimony and witness information gathering sessions, but the deliberative process is conducted in closed session.  This approach may be suitable for preparing significant policy decisions by public institutions and/or governments.

Danish Consensus Conference is a participatory model that enjoins citizens to consider controversial STEM issues and report their policy recommendations to the government and the public.  The goal is to form a consensus that strikes a balance between competing stakeholder demands, known risks and potential harms to society, and economic benefits.  This framework is particularly apt for public assessment of large, high tech systems for the purpose of policy guidance.  Nonetheless, it is not a model familiar to most American stakeholders and may not be appropriate for all AI-in-society deliberations.

Indigenous Data Sovereignty Governance (IDSG) is an emerging framework situated in the human rights principle of self determination.  It is an ethical process that calls on place-based community standards, norms, and traditions.  It can entail exceptionally time consuming deliberation that results in the establishment of long-term, communitywide rules.

Institutional Review Board (IRB) is a well-established expert-based committee process that exists primarily to reduce risk and limit exposure to the institution by confirming that the plans and processes involved in the activity comply with laws and regulations.  It extends protections to individuals and communities indirectly by restricting activities to the confines of broadly applicable legal rules.  This approach may be particularly suitable for controversial policy actions, where watchdog organizations are viewed as adversarial actors, rather than community partners.

Jury Deliberation is a type of peer review that allows oversight entities to review evidence and ponder narrow questions in private or closed sessions.  Jury decisions must be unanimous and can be appealed.  Juries are not required to justify their decisions.  This approach might be preferable when sensitive data is involved.

Situated Learning/Community Management (COP) is a Community of Practice (COP) framework that fosters peer-based skilling, adoption of best practices, co-creation of  design and development, and information dissemination.  This is likely to be a strong mechanism for project planning and early-stage implementation of public AI systems.

Truth and Reconciliation Commission (TRC) is an arbiter-led approach that allows community actors to register grievances, state their trustworthy goals, define for themselves the role that they want to play in the design or oversight of trustworthy AI systems, and reset with a restorative justice engagement framework going forward.  

Unstructured Athenian (UA) model, which provides stakeholders a pathway to identify bias outcomes and issue tickets for resolution by administrators within a predetermined timeframe.  This is likely to appeal to communities working in close partnership with AI stakeholders, for example where large data centers are welcomed and/or AI systems are perceived as meeting a defined need in service to the local community.

Phase II: Pilot Studies

Phase II will test known frameworks that have been extensively studied, documented, reviewed, and refined with proven best practices over years and even decades.  These reference frameworks represent a variety of input and deliberation processes that communities can select according to needs and capacities.  

Volunteers will select an oversight mechanism of their choice and join other members of the group to review a real use case that is relevant to their competency and/or community.  The goal is to study the effectiveness of the process itself, not the individual or group dynamics, per se.  For example, an oversight goal might include “To determine the suitability of Flock Safety devices in a defined rural geolocation.”  The pilot group will be organized according to the standards of their assigned oversight mechanism, e.g., Jury, and given instructions for deliberation.  Each volunteer will complete a survey instrument designed to capture metrics about the effectiveness of the process in the context of assessing AI in society, weaknesses and strengths of the mechanism to deliver the desired goal(s), and general feedback to describe flaws and suggest improvements.

Phase III: Development of AI Oversight Mechanisms

The goal of the Phase III is twofold: (1) to apply findings from the set of reference frameworks toward the development of public oversight mechanisms that are suitable for monitoring AI in society; and (2) development of decision tooling that will enable the lay public to choose an appropriate framework that fits the specific need, capacity, competency, and context.  It is entirely possible that an AI tool will be implemented in the development of this tooling.

Phase IV: Findings and Recommendations

While reflexion is built into the methodology and data will be produced and analyzed throughout the study, the pilot phase will conclude after at least one set of metrics and survey instruments has been collected for each reference framework.  A preliminary report will be produced to report findings with the option for further investigation and framework development, as warranted.

Disclosure Statements

AI

The textual content of this proposal was drafted in Google Docs without AI assistance.  The visual content (i.e., the infographic) was created using Google Gemini.

Open Access

This project has been pre-registered with OSF under DOI: https://osf.io/zf65u.

References

Aliaga, Stephanie. “How Much of the AI Boom Is Adding to U.S. Growth?” Jpmorgan.com. J.P. Morgan Asset Management, September 2, 2026. https://am.jpmorgan.com/us/en/asset-management/adv/insights/market-insights/market-updates/on-the-minds-of-investors/how-much-of-the-ai-boom-is-adding-to-us-growth/. 

Ashery, Ariel Flint, Luca Maria Aiello, and Andrea Baronchelli. “Emergent Social Conventions and Collective Bias in LLM Populations.” Science Advances 11, no. 20 (May 16, 2025). https://doi.org/10.1126/sciadv.adu9368. 

Australian Cyber Security Centre. “Defending against AI-enabled cyber attacks: Guidance for government, critical infrastructure and large enterprises: Recommended actions for government, critical infrastructure providers and large enterprises to defend against AI-enabled cyber attacks.” Poster. July 6, 2026. Australian Signals Directorate. https://www.cyber.gov.au/publication/defending-against-ai-enabled-cyber-attacks-guidance-for-government-critical-infrastructure-and-large-enterprises. 

Brockie, Teresa N., Kyle Hill, Patricia M. Davidson, Ellie Decker, Lydia Koh Krienke, Katie E. Nelson, Natalie Nicholson, Alicia M. Werk, Deborah Wilson, and Deana Around Him. 2022. “Strategies for Culturally Safe Research with Native American Communities: An Integrative Review.” Contemporary Nurse 58 (1): 8–32. Doi: 10.1080/10376178.2021.2015414. 

Carroll SR, Garba I, Plevel R, Small-Rodriguez D, Hiratsuka VY, Hudson M and Garrison NA (2022) Using Indigenous Standards to Implement the CARE Principles: Setting Expectations through Tribal Research Codes. Front. Genet. 13:823309. doi: 10.3389/fgene.2022.823309

Davidson, Tom, Halperin, Basil, Houlden, Thomas and Korinek, Anton. “When Does Automating AI Research Produce Explosive Growth? Feedback Loops in Innovation Networks.” NBER Working Paper No. 35155. April 2026. https://www.nber.org/system/files/working_papers/w35155/w35155.pdf. 

Eisenhammer, Stephen. “AI’s Race to Transform the World before the Money Runs Out.” Reuters, October 3, 2026. ttps://www.reuters.com/business/retail-consumer/ais-race-transform-world-before-money-runs-out-2026-10-03/. 

Flock Safety. “Use Cases.” Flocksafety.com, 2026. https://www.flocksafety.com. 

Garba, Ibrahim and Stephanie Russo Carroll, “ The UN Declaration on the Rights of Indigenous Peoples and Genomics: Ethical Complementarity for Just Research,” in “Envisioning a More Just Genomics,” ed. Josephine Johnston, Deanne Dunbar Dolan, Danielle M. Pacia, Sandra Soo-Jin Lee, and Mildred K. Cho, special report, Hastings Center Report 54, no. S2 (2024): S120–S125. DOI: 10.1002/hast.4937 

Goudarzi, Sara. “How the Bad Science of AI Doomerism Is Good for Big Business.” Bulletin of the Atomic Scientists, September 30, 2026. https://thebulletin.org/2026/09/how-the-bad-science-of-ai-doomerism-is-good-for-big-business/. 

  Joss, Simon. Danish consensus conferences as a model of participatory technology assessment: An impact study of consensus conferences on Danish Parliament and Danish public debate, Science and Public Policy, Volume 25, Issue 1, February 1998, Pages 2–22, https://doi.org/10.1093/spp/25.1.2 

Kim, Victoria. “OpenAI Says It Changed Systems After Australia Hacking.” The New York Times, October 6, 2026. https://www.nytimes.com/2026/10/05/world/australia/australia-openai-hearing-breach.html. 

Klein, Ezra and David Robinson. “‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit. David Robinson, who wrote OpenAI’s safety protocols, resigned last week over safety concerns.” Interview transcript. The New York Times, October 8, 2026. https://www.nytimes.com/video/opinion/100000011191541/this-is-nuts-an-openai-insider-explains-why-he-quit.html. 

Perko, Tanja. 2025. "The Danish Consensus Conference: A Critical Evaluation," Risk, Governance and Society, in: Fairness and Competence in Citizen Participation, pages 189-212, Springer.

Phillips, P, Theodore Jensen, Patrick Hall, Razvan Amironesei, Yee-Yin Choong, Craig Greenberg, and Kristen Greene. “The TEVV-Athlon Framework for NIST Trustworthy and Responsible AI.”  NIST AI 200-2 Ipd (Initial Public Draft). August 2026. https://doi.org/10.6028/NIST.AI.200-2.ipd. ‌

Romm, Tony. “Trump Shrugs Off A.I. And Data Center Concerns as Voters Grow Anxious.” nytimes. The New York Times, August 10, 2026. https://www.nytimes.com/2026/08/10/business/trump-artificial-intelligence-data-centers-ai.html.‌ 

  Sharan, Malvika, Emma Karoune, Vicky Hellon, Cassandra Gould van Praag, Kayumbi Gabin, Arielle Bennett, Alexandra Araujo Alvarez, et al. “Professionalising Community Management Roles in Interdisciplinary Research Projects,” August 27, 2024. arXiv. https://arxiv.org/abs/2409.00108v1. 

  United Nations. “United Nations Declaration on the Rights of Indigenous Peoples.” United Nations Declaration On The Rights Of Indigenous Peoples, September 13, 2007. https://doi.org/10.1353/hrq.2011.0040.  

US Bureau of Economic Analysis. “Gross Domestic Product.” BEA.gov, 2026. https://www.bea.gov/data/gdp/gross-domestic-product.