Legal Help Commons

JusticeBench

Explore What Is Being Built in Justice AI

Discover AI projects across the access to justice landscape, understand where they fit in the justice journey, and find data and evaluation tools to know if they work. JusticeBench is the discovery layer of the Legal Help Commons.

The JusticeBench platform is under development. Please give us your feedback and ideas! Share feedback & projects

Projects

What kinds of AI projects are already in the works to advance access to justice? Filter by issue area or project status to find what’s relevant to your work.

Showing 59 of 59 projects

Consumer Legal Services

Consumer legal services help people with money and financial problems, to do things like responding to a debt collection lawsuit, dealing with wage or bank garnishments, stopping unfair collections, correct credit errors, and recover from scams or abusive lending.

1 project

Education Legal Services

Education legal services help students and their families by resolving special education, discipline, and enrollment issues through advocacy, hearings, and services coordination.

1 project

Estates

This category covers planning for end-of-life, possible incapacitation, and other special circumstances that would prevent a person from making decisions about their own well-being, finances, and property. This includes issues around wills, powers of attorney, advance directives, trusts, guardianships, conservatorships, and other estate issues that people and families deal with. Scenarios in this category include making or changing a will/living will/advance directive, setting up a trust or power of attorney, and help with a probate or administering an estate.

1 project

Public Benefits Legal Services

Public benefits legal services help people access food, cash, health, and disability benefits, including by screening eligibility, assembling applications, and appealing denials.

1 project

Know of a project not listed here? Tell us about it.

Data & Evaluation

This section collects datasets, benchmarks, evaluation protocols, taxonomies, and leaderboards for building and testing legal AI. Use these to train models, measure performance, and establish quality standards.

Test & Evaluate

Use benchmarks, eval Q&A, and test suites to score model performance.

LegalBench benchmark tasks

LegalBench benchmark tasks

Test & Evaluate

A collaboratively built suite of 160+ legal tasks that measure legal reasoning across statutes, cases, contracts, and procedures—crafted by legal experts and useful to test the performance of models and solutions.

Beagle+ Legal Chatbot Testing Dataset

Beagle+ Legal Chatbot Testing Dataset

Test & Evaluate

A curated, labeled dataset of 42 real-world legal question-answers used to evaluate the safety and helpfulness of AI-generated answers during the development of Beagle+, a legal information chatbot for British Columbia.

OpenAI GDPval test on lawyer tasks

OpenAI GDPval test on lawyer tasks

Test & Evaluate

A short set of evaluation + dataset resources from OpenAI that tests AI models on economically valuable, job-specific tasks—including a “Professional: Lawyers” track with realistic legal work artifacts and human grading.

PRBench

PRBench

Test & Evaluate

PRBench is a public, expert-authored Law and Finance benchmark that evaluates models on open-ended professional tasks using detailed, weighted rubrics (10–30 criteria per task) and reports fine-grained performance, including on a designated “Hard” subset.

User Intake Classification labeled dataset

User Intake Classification labeled dataset

Test & Evaluate

A 180-item labeled dataset of user and provider intake/proflile inputs, with jurisdiction, legal issue LIST codes, emergency flags, and audience tags. It can be used to evaluate systems to extract structured case data from unstructured legal-help intake. Covers 37 states, 8 input channels, the full Legal Help Commons audience-served vocabulary, and negative test cases for robustness measurement.

Build & Train

Use labeled data, training corpora, and synthetic queries to build tools.

LIST-Labeled Legal Q Set (L3Q)

LIST-Labeled Legal Q Set (L3Q)

Build & Train

L3Q is a synthetic dataset of 3,300+ legal help questions and search prompts, labeled with LIST issue codes, jurisdiction/entity tags, and question sophistication, available in Airtable from Legal Design Lab.

Common Legal Help Questions

Common Legal Help Questions

Build & Train

A curated, synthetic dataset of common civil legal questions asked by the public—tagged by legal issue and sensitivity—to support research, product testing, and development of legal help technologies.

Learned Hands: Labeled Dataset of Legal Issues in Reddit problem stories

Learned Hands: Labeled Dataset of Legal Issues in Reddit problem stories

Build & Train

A crowdsourced, expert-reviewed dataset of legal issue labels on real-world problem narratives, created to support machine learning and research on legal needs and access to justice.

Common-48-Legal-Queries

Common-48-Legal-Queries

Build & Train

A set of 48 legal help common questions, all synthetic, tied to a jurisdiction, with 2-3 sentences of context.

Legal Help Synthetic Query Pack - LHSQ115

Legal Help Synthetic Query Pack - LHSQ115

Build & Train

A set of 115 fully synthetic legal-help queries drafted from patterns observed in 2025 online legal help questions from members of the US public, labeled with LIST issue codes, language, and key metadata for safe sharing and evaluation.

High Risk Legal Help Queries

High Risk Legal Help Queries

Build & Train

User queries designed to catch failure modes, across high-stakes legal scenarios in six U.S. states.

CourtListener

CourtListener

Build & Train

A free, public database of U.S. court opinions, federal court filings, judges, and oral arguments, maintained by Free Law Project. In operation since 2010.

Please share datasets with JusticeBench at this form.

Guides

How can you create an AI plan for your justice organization, and what's the best way to implement new AI developments? Explore our guides for justice institution leaders.

Legal Aid AI Implementation guide

Legal Aid AI Implementation guide↗

A guide from A2J Tech and LSNTAP on how to implement AI responsibly in a legal aid group.

AI For Legal Aid: a Step-By-Step Guide

AI For Legal Aid: a Step-By-Step Guide↗

A brief gameplan to help legal aid leaders identify opportunities, choose the right tools and partners, and launch a new AI solution.

Innovating for Access: AI-Enhanced Triage & Intake for Legal Services Organizations

Innovating for Access: AI-Enhanced Triage & Intake for Legal Services Organizations↗

The paper describes results of 2025 interviews with legal professionals and technologists to find out how organizations are actually using AI in their triage and intake systems. It maps what's being built, what's working, and what organizations need to have in place before they start.

ABA Ethics formal opinion 512 on GenAI

ABA Ethics formal opinion 512 on GenAI↗

Guidance from the American Bar Association on how lawyers can responsibly adopt AI tools while meeting professional conduct obligations.

AI Readiness for State Courts

AI Readiness for State Courts↗

A practical NCSC guide that helps state and local courts assess and improve their AI readiness. It combines a clear maturity model (Foundations → Implementation → Post-project feedback), lifecycle guidance, and an interactive self-assessment to plan, govern, and evaluate AI responsibly.

AI Risk Management Framework for Legal Teams

AI Risk Management Framework for Legal Teams↗

A practical, step-by-step framework from Duke RAILS to help legal teams identify, govern, and mitigate AI-related risks while supporting responsible innovation.

Audit report for Nevada Courts Self-Help Chatbot

Audit report for Nevada Courts Self-Help Chatbot↗

This audit report investigates a self-help chatbot designed for Nevada litigants with family law or probate cases. The audit was conducted in Spring 2026 as part of Duke’s AI Audits and Access to Justice practicum, in partnership with the Nevada Administrative Office of the Courts.

Cal. State Bar Practical Guide for use of AI

Cal. State Bar Practical Guide for use of AI↗

Ethical framework from the California State Bar guiding lawyers through the responsible use of generative AI in legal practice.

LIA chatbot evaluation by Duke Law

LIA chatbot evaluation by Duke Law↗

A report of an independent audit of the LIA chatbot in North Carolina by Duke Law researchers, that documents an evaluation protocol, usage results, issues to address, and recommendations for improvement.

LSC's AI Peer Learning Labs

LSC's AI Peer Learning Labs↗

An online hub & regular online sessions where civil legal aid programs share lessons, resources, and best practices for using AI responsibly in legal services.

NCSC Guides for Implementing AI in Courts

NCSC Guides for Implementing AI in Courts↗

Guidance from the National Center for State Courts and partners on adopting AI in court operations responsibly, with clear policies, oversight, and safeguards.

Regulating AI in the Delivery of Consumer-Facing Legal Services

Regulating AI in the Delivery of Consumer-Facing Legal Services↗

IAALS' guide to a phased, innovation-friendly approach to regulating AI tools that deliver legal information and services directly to the public.

Please share guide proposals and open-source materials with us at this form.