Shaping the Future of AI for Access to Justice

By Margaret Hagan, originally published on Legal Design & Innovation

If AI is going to advance access to justice rather than deepen the justice gap, the public-interest legal field needs more than speculation and pilots — we need statewide stewardship.

We need specific people and institutions in every state who wake up each morning responsible for two things:

1. AI readiness and vision for the legal services ecosystem: getting organizations knowledgeable, specific, and proactive about where AI can responsibly improve outcomes for people with legal problems — and improve the performance of services. This can ensure the intelligent and impactful adoption of AI solutions as they are developed.
2. AI R&D encouragement and alignment: getting vendors, builders, researchers, and benchmark makers on the same page about concrete needs; matchmaking them with real service teams; guiding, funding, evaluating, and communicating so the right tools get built and adopted.

Ideally, these local state stewards will be talking with each other regularly. In this way, there can be federated research & development of AI solutions for legal service providers and the public struggling with legal problems.

This essay outlines what AI + Access to Justice stewardship could look like in practice — who can play the role, how it works alongside court help centers and legal aid, and the concrete, near-term actions a steward can take to make AI useful, safe, and truly public-interest.

Why stewardship — why now?

Every week, new tools promise to draft, translate, summarize, triage, and file. Meanwhile, most legal aid organizations and court help centers are still asking foundational questions: What’s safe? What’s high-value? What’s feasible with our staff and privacy rules? How do we avoid vendor lock-in? How do we keep equity and client dignity at the center?

Without stewardship, AI adoption will be fragmented, extractive, and inequitable. With stewardship, states can:

• Focus AI where it demonstrably helps clients and staff. Prioritize tech based on community and provider stakeholders’ needs and preferences — not just what is being sold by vendors.
• Prepare data and knowledge so tools work in the local contexts. Also, that they can be trained safely & benchmarked responsibly with relevant data that is masked and safe.
• Align funders, vendors, and researchers around real service needs. So that all of these stakeholder groups, with their capacity to support, build, and evaluate emerging technology, direct this capacity at opportunities that are meaningful.
• Develop shared evaluation and governance so we build trust, not backlash.

Who can play the Statewide AI Steward role?

“Steward” is a role, not a single job title. Different kinds of groups can carry it, depending on how your state is organized:

• Access to Justice Commissions / Bar associations / Bar foundations that convene stakeholders, fund statewide initiatives, and set standards.
• Legal Aid Executive Directors (or cross-org consortia) with authority to coordinate practice areas and operations.
• Court innovation offices / judicial councils that lead technology, self-help, and rules-of-court implementations.
• University labs / legal tech nonprofits that have capacity for research, evaluation, data stewardship, and product prototyping.
• Regional collaboratives with a track record of shared infrastructure and implementation.

Any of these can steward. The common denominator: local trusted relationships, coordination power, and delivery focus. The steward must be able to convene local stakeholders, communicate with them, work with them on shared training and data efforts, and move from talk to action.

The steward’s two main missions

Mission 1: AI readiness + vision (inside the legal ecosystem)

The steward gets legal organizations — executive directors, supervising/managing attorneys, practice leads, intake supervisors, operations staff — knowledgeable and specific about where AI can responsibly improve outcomes. This means:

• Translating AI into service-level opportunities (not vague “innovation”).
• Running short, targeted training sessions for leaders and teams.
• Co-designing workflow pilots with clear review and safety protocols.
• Building a roadmap: which portfolios, which tools, what sequence, what KPIs.
• Clarify ethical, privacy, and consumer/client safety priorities and strategies, to talk about risks and worries in specific, technically-informed ways that provide sufficient protection to users and orgs — and don’t fall into inaction because of ill-defined concern about risk.

The result: organizations are in charge of the change rather than passive recipients of vendor pitches or media narratives.

2) AI tech encouragement + alignment (across the supply side)

The steward gets the groups who specialize in building and evaluating technology — vendors, tech groups, university researchers, benchmarkers— pointed at the right problems with the right real-world partnerships:

• Publishing needs briefs by portfolio (housing, reentry, debt, family, etc).
• Matchmaking teams and vendors; structuring pilots with data, milestones, evaluation, and governance. Helping organizations choose a best-in-class vendor and then also manage this relationship with regular evaluation.
• Contributing to benchmarks, datasets, and red-teaming so the field learns together. Build the infrastructure that can lead to effective, ongoing evaluation of how AI systems are performing.
• Helping fund and scale what works; communicating results frankly. Ensuring that prototypes and pilots’ outcomes are shared to inform others of what they might adopt, or what changes must happen to the AI solutions for them to be adopted or scaled.

The result: useful and robust AI solutions built with frontline reality, evaluated transparently, and ready to adopt responsibly.

What Stewards Could Do Month-to-Month

I have been brainstorming specific actions that a statewide steward could do. Many of these actions could also be done in concert with a federated network of stewards.

Map the State’s Ecosystem of Legal Help

Too often, we think in terms of organizations — “X Legal Aid,” “Y Court Help Center” — instead of understanding who’s doing the actual legal work.

Each state needs to start by identifying the legal teams operating within its borders.

• Who is doing eviction defense?
• Who helps people with no-fault divorce filings?
• Who handles reasonable accommodation letters for tenants?
• Who runs the reentry clinic or expungement help line?
• Who offers debt relief letter assistance?
• Who does restraining order help?

This means mapping not just legal help orgs, but service portfolios and delivery models. What are teams doing? What are they not doing? And what are the unmet legal needs that clients consistently face?

This is a service-level analysis — an inventory of the “market” of help provided and the legal needs not yet met.

AI Training for Leaders + Broader Legal Organizations

Most legal aid and court help staff are understandably cautious about AI. Many don’t feel in control of the changes coming — they feel like they’re watching the train leave the station without them.

The steward’s job is to change that.

• Demystify AI: Explain what these systems are and how they can support (or undermine) legal work.
• Coach teams: Help practice leads and service teams see which parts of their work are ripe for AI support.
• Invite ownership: Position AI not as a threat, but as a design space — a place where legal experts get to define how tools should work, and where lawyers and staff retain the power to review and direct.

To do this, stewards can run short briefings for EDs, intake leads, and practice heads on LLM basics, use cases, risks, UPL and confidentiality, and adoption playbooks. Training aims to get them conversant in the basics of the technology and help them envision where responsible opportunities might be. Let them see real-world examples of how other legal help providers are using AI behind the scenes or directly to the public.

Brainstorm + Opportunity Mapping Workshops with Legal Teams

Bring housing teams, family law facilitator teams, reentry teams, or other specific legal teams together. Have them map out their workflows and choose which of their day-to-day tasks is AI-opportune. Which of the tasks are routine, templated, and burdensome?

As stewards run these workshops, they can be on the lookout for where legal teams in their state can build, buy, or adopt an AI solution in 3 areas.

Brainstorm 1: AI Copilots for Services Legal Teams Already Offer

This is the lowest-risk, highest-benefit space. Legal teams are already helping with eviction defense, demand letters, restraining orders, criminal record clearing, etc.

Here, AI can act as a copilot for the expert — a tool that does things that the expert lawyer, paralegal, or legal secretary is already doing in a rote way:

• Auto-generates first drafts based on intake data
• Summarizes client histories
• Auto-fills court forms
• Suggests next actions or deadlines
• Creates checklists, declarations, or case timelines

These copilots don’t replace lawyers. They reduce drudge work, improve quality, and make staff more effective.

Brainstorm 2: AI Copilots for Services That Could Be Done by Pro Bono or Volunteers

Many legal aid organizations know where they could use more help: limited-scope letters, form reviews, answering FAQs, or helping users navigate next steps.

AI can play a key role in unlocking pro bono, brief advice, and volunteer capacity:

• Automating burdensome tasks like collecting or review database records,
• Helping them write high-quality letters or motions
• Pre-filling petitions and forms with data that has been gathered
• Providing them with step-by-step guidance
• Flagging errors, inconsistencies, or risks in drafts
• Offering language suggestions or plain-language explanations

Think of this as AI-powered “training wheels” that help volunteers help more people, with less handholding from staff.

Brainstorm 3: AI Tools for Services That Aren’t Currently Offered — But Should Be

There are many legal problems where there is high demand, but legal help orgs don’t currently offer help because of capacity limits.

Common examples of these under-served areas include:

• Security deposit refund letters
• Creating demand letters
• Filing objections to default judgments
• Answering brief questions

In these cases, AI systems — carefully designed, tested, and overseen — can offer direct-to-consumer services that supplement the safety net:

• Structured interviews that guide users through legal options
• AI-generated letters/forms with oversight built in
• Clear red flags for when human review is needed

This is the frontier: responsibly extending the reach of legal help to people who currently get none. The brainstorm might also include reviewing existing direct-to-consumer AI tools from other legal orgs, and deciding which they might want to host or link to from their website.

The steward can hold these brainstorming and prioritization sessions to help legal teams find these legal team co-pilots, pro bono tools, and new service offerings in their issue area. The stewards and legal teams can move the AI vision forward & prepare for a clear scope for what AI should be built.

Data Readiness + Knowledge Base Building

Work with legal and court teams to inventory what data they have that could be used to train or evaluate some of the legal AI use cases they have envisioned. Support them with tools & protocols by which to mask PII in this document and make it safe to use in AI R&D.

This could mean getting anonymized completed forms, documents, intake notes, legal answers, data reports, or other legal workflow items. Likely, much of this data will have to be labeled, scored, and marked up so that it’s useful in training and evaluation.

The steward can help the groups that hold this data to understand what data they hold, how to prepare it and share it, and how to mark it up with helpful labels.

Part of this is also to build a Local Legal Help Knowledge Base — not just about the laws and statutes on the books, but about the practical, procedural, and service knowledge that people need when trying to deal with a legal problem.

Much of this knowledge is in legal aid lawyers’ and court staff’s heads, or training decks and events, or internal knowledge management systems and memos.

Stewards can help these local organizations contribute this knowledge about local legal rules, procedures, timelines, forms, services, and step-by-step guides into a statewide knowledge base. This knowledge base can then be used by the local providers. It will be a key piece of infrastructure on which new AI tools and services can be built.

Adoption Logistics

As local AI development visions come together, the steward can lead on adoption logistics.

The steward can make sure that the local orgs don’t reinvent what might already exist, or spend money in a wasteful way.

They can do tool evaluations to see which LLMs and specific AI solutions perform best on the scoped tasks. They can identify researchers and evaluators to help with this. They can also help organizations procure these tools or even create a pool of multiple organizations with similar needs for a shared procurement process.

They might also negotiate beneficial, affordable licenses or access to AI tools that can help with the desired functions. They can also ensure that case management and document management systems are responsive to the AI R&D needs, so that the legacy technology systems will integrate well with the new tools.

Ideally, the steward will help the statewide group and the local orgs make smart investments in the tech they might need to buy or build — and can help clear the way when hurdles emerge.

Bigger-Picture Steward Strategies

In addition to these possible actions, statewide stewards can also follow a few broader strategies to get a healthy AI R&D ecosystem in their state and beyond.

Be specific to legal teams

As I’ve already mentioned throughout this essay, stewards should be focused on the ‘team’ level, rather than the ‘organization’ one. It’s important that they develop relationships and run activities with teams that are in charge of specific workflows — and that means the specific kind of legal problem they help with.

Stewardship should be organizing its statewide network of named teams and named services, for example,

• Housing law teams & their workflows: hotline consults, eviction defense prep, answers, motions to set aside, trial prep, RA letters for habitability issues, security-deposit demand letters.
• Reentry teams & their workflows: record clearance screening, fines & fees relief, petitions, supporting declarations, RAP sheet interpretation, collateral consequences counseling.
• Debt/consumer teams & their workflows: answer filing, settlement letters, debt verification, exemptions, repair counseling, FDCPA dispute letters.
• Family law teams & their workflows: form prep (custody, DV orders), parenting plans, mediation prep, service and filing instructions, deadline tracking.

The steward can make progress on its 2 main goals — AI readiness and R&D encouragement — if it can build a strong local network among the teams that work on similar workflows, with similar data and documents, with similar audiences.

Put ethics, privacy, and operational safeguards at the center

Stewardship builds trust by making ethics operational rather than an afterthought. This all happens when AI conversations are grounded, informed, and specific among legal teams and communities. It also happens when they work with trained evaluators, who know how to evaluate the performance of AI rigorously, not based on anecdotes and speculation.

The steward network can help by planning out and vetting common, proven strategies to ensure quality & consumer protection are designed into the AI systems. They could work on:

• Competence & supervision protocols: helping legal teams plan for the future of expert review of AI systems, clarifying “eyes-on” review models with staff trainings and tools. Stewards can also help them plan for escalation paths, when human reviewers find problems with the AI’s performance. Stewards might also work on standard warnings, verification prompts, and other key designs to ensure that reviewers are effectively watching AI’s performance.
• Professional ethics rules clarity: help the teams design internal policies that ensure they’re in compliance with all ethical rules and responsibilities. Stewards can also help them plan out effective disclosures and consent protocols, so consumers know what is happening and have transparency.
• Confidentiality & privacy: This can happen at the federated/ national level. Stewards can set rules for data flows, retention, de-identification/masking — which otherwise can be overwhelming for specific orgs. Stewards can also vet vendors for security and subprocessing.
• Accountability & Improvements: Stewards can help organizations and vendors plan for good data-gathering & feedback cycles about AI’s performance. This can include guidance on document versioning, audit logs, failure reports, and user feedback loops.

Stewards can help bake safeguards into workflows and procurement, so that there are ethics and privacy by design in the technical systems that are being piloted.

Networking stewards into a federated ecosystem

For statewide stewardship to matter beyond isolated pilots, stewards need to network into a federated ecosystem — a light but disciplined network that preserves local autonomy while aligning on shared methods, shared infrastructure, and shared learning.

The value of federation is compounding: each state adapts tools to local law and practice, contributes back what it learns, and benefits from the advances of others. Also, many of the tasks of a steward — educating about AI, building ethics and safeguards, measuring AI, setting up good procurement — will be quite similar state-to-state. Stewards can share resources and materials to implement locally.

What follows reframes “membership requirements” as the operating norms of that ecosystem and explains how they translate into concrete habits, artifacts, and results.

Quarterly check-ins become the engine of national learning. Stewards participate in a regular virtual cohort, not as a status ritual but as an R&D loop. Each session surfaces what was tried, what worked, and what failed — brief demos, before/after metrics, and annotated playbooks.

Stewards use these meetings to co-develop materials, evaluation rubrics, funding strategies, disclosure patterns, and policy stances, and to retire practices that didn’t pan out. Over time, this cadence produces a living canon of benchmarks and templates that any newcomer steward can adopt on day one.

Each year, the steward could champion at least one pilot or evaluation (for example, reasonable-accommodation letters in housing or security-deposit demand letters in consumer law), making sure it has clear success criteria, review protocols, and an exit ramp if risks outweigh benefits. This can help the pilots spread to other jurisdictions more effectively.

Shared infrastructure is how federation stays interoperable. Rather than inventing new frameworks in every state, stewards lean on common platforms for evaluation, datasets, and reusable workflows. Practically, that means contributing test cases and localized content, adopting shared rubrics and disclosure patterns, and publishing results in a comparable format.

It also means using common identifiers and metadata conventions so that guides, form logic, and service directories can be exchanged or merged without bespoke cleanup. When a state localizes a workflow or improves a safety check, it pushes the enhancement upstream, so other states can pull it down and adapt with minimal effort.

Annual reporting turns stories into evidence and standards. Each steward could publish a concise yearly report that covers: progress made, obstacles encountered, datasets contributed (and their licensing status), tools piloted or adopted (and those intentionally rejected), equity and safety findings, and priorities for the coming year.

Because these reports follow a common outline, they are comparable across states and can be aggregated nationally to show impact, surface risks, and redirect effort. They also serve as onboarding guides for new teams: “Here’s what to try first, here’s what to avoid, here’s who to call.”

Success in 12–18 months looks concrete and repeatable. In a healthy federation, we could point to a public, living directory of AI-powered teams and services by portfolio, with visible gaps prioritized for action.

• We could have several legal team copilots embedded in high-volume workflows — say, demand letters, security-deposit letters, or DV packet preparation — with documented time savings, quality gains, and staff acceptance.
• We could have volunteer unlocks, where a clinic or pro bono program helps two to three times more people in brief-service matters because a copilot provides structure, drafting support, and review checkers.
• We could have at least one direct-to-public workflow launched in a high-demand, manageable-risk area, with clear disclosures, escalation rules, and usage metrics.
• We would see more contributions to data-driven evaluation practices and R&D protocols. This could be localized guides, triage logic, form metadata, anonymized samples, and evaluation results. Or it could be an ethics and safety playbook that is not just written but operationalized in training, procurement, and audits.

A federation of stewards doesn’t need heavy bureaucracy. It could be a set of light, disciplined habits that make local work easier and national progress faster. Quarterly cohort exchanges prevent wheel-reinventing. Local duties anchor AI in real services. Shared infrastructure keeps efforts compatible. Governance protects the public-interest character of the work. Annual reports convert experience into standards.

Put together, these practices allow stewards to move quickly and responsibly — delivering tangible improvements for clients and staff while building a body of knowledge the entire field can trust and reuse.

Stewardship as the current missing piece

Our team at Stanford Legal Design Lab is aiming for an impactful, ethical, robust ecosystem of AI in legal services. We are building the platform JusticeBench to be a home base for those working on AI R&D for access to justice. We are also building justice co-pilots directly with several legal aid groups.

But to build this robust ecosystem, we need local stewards for state jurisdictions across the country — who can take on key leadership roles and decisions — and make sure that there can be A2J AI that responds to local needs but benefits from national resources. Stewards can also help activate local legal teams, so that they are directing the development of AI solutions rather than reacting to others’ AI visions.

We can build legal help AI state by state, team by team, workflow by workflow. But we need stewards who keep clients, communities, and frontline staff at the center, while moving their state forward.

That’s how AI becomes a force for justice — because we designed it that way.