A pilot program in Colorado tested an AI-assisted tool designed to help unemployment benefits adjudicators ask better questions and reach faster decisions. The tool, built by Stanford Law School's Regulation, Evaluation, and Governance Lab (RegLab) and the Colorado Department of Labor and Employment (CDLE), received an inaugural AI in Action Award from the Center for Civic Futures in the "Rising AI Pilot" category.
The award recognizes practical uses of artificial intelligence in government that prioritize responsible adoption and shareable lessons across jurisdictions. The Center for Civic Futures is a nonpartisan nonprofit focused on responsible AI and emerging technology in government. RegLab also earned an honorable mention for separate work with the Santa Clara County Office of the Clerk-Recorder on a racial covenant identification tool.
What the tool does
Adjudicators handling unemployment claims face a familiar bottleneck: determining which facts matter, spotting what is missing or disputed, and deciding what to ask next - all while managing heavy caseloads. The RegLab-CDLE tool keeps human judgment at the center. It guides an adjudicator from the specific facts of a claim to relevant follow-up topics and questions, rather than replacing the decision-maker.
The design reflects RegLab's focus on the intersection of law, data science, and public policy. The goal was not automation for its own sake, but a structured assistant that helps a trained professional move through a complex fact pattern more efficiently.
Evidence over anecdote
What set this project apart was its commitment to rigorous evaluation. The team ran a randomized controlled trial using real adjudicators and historical cases, then published the findings so other government agencies could learn from the results. The Center for Civic Futures pointed to that evidence base as a distinguishing factor.
"Government agencies are being asked to solve extraordinarily hard problems, from delivering benefits to protecting public health and the environment, often without access to the scientific and technical capacity those challenges require," said Daniel E. Ho, faculty director of RegLab and associate director of Stanford's Institute for Human Centered AI. "This project shows what can happen when government and academic partners work together to bring modern data science to core public systems in ways that are careful, measurable, and grounded in the realities of public service."
Amanda Neal, Strategic Business Technology Director at CDLE, said, "It's wonderful to see this Colorado-RegLab collaboration around responsible AI for benefits adjudication recognized. We are excited about this model of responsible AI to modernize a system that provides critical support for so many."
How the work fits into a larger effort
The Colorado pilot is one piece of RegLab's broader collaboration-based approach to modernizing government. The lab develops partnerships with agencies and nonprofits to run demonstration projects in core governance functions - including public benefits, tax administration, regulatory compliance, and AI governance - while strengthening public-sector capacity. The work was supported in part by Arnold Ventures, Stanford Impact Labs, and Stanford HAI.
RegLab's AI assistance work in benefits adjudication has also received recent recognition at the CS&Law Symposium and the International Conference on Law and AI. For legal professionals working in or adjacent to government systems, these projects offer a concrete model of how AI for Legal applications can be introduced into high-stakes administrative processes without displacing the human judgment that final decisions require. The same principles apply broadly to AI for Government initiatives where accuracy, fairness, and speed are all non-negotiable.
Why this matters for legal professionals
For lawyers, hearing officers, and compliance specialists who work on benefits adjudication or administrative law, the Colorado pilot signals a shift in what responsible AI deployment looks like. The tool did not replace the adjudicator; it structured the fact-finding process. The team tested it under controlled conditions and published the data. That combination - domain-specific design, human-in-the-loop operation, and published evaluation - provides a template for how legal workflows can integrate machine assistance without sacrificing due process or accountability. When a tool is built to surface the next relevant question rather than to recommend an outcome, it fits more naturally into the reasoning that legal professionals already perform.
Your membership also unlocks:
