We are happy to announce the launch of our fourth round of the class “AI for Legal Help”. It is cross-listed at Stanford Law School and Design School.
Students will be working with real-world, public interest legal groups to develop AI solutions in a responsible, practical way — that can help scale out high-need legal services.
Here is the class description:
Want to build AI that actually matters? AI for Legal Help is a two-quarter, hands-on course where law, design, computer science, and policy students team up with legal aid organizations and court self-help centers to take on one of the biggest challenges in tech today: using AI to expand access to justice.
You’ll work directly with real-world partners to uncover where AI could make legal services faster, more scalable, and more effective—while ensuring it’s safe, ethical, and grounded in the realities of public service. From mapping workflows to spotting opportunities, from creating benchmarks and datasets to designing AI “co-pilots” or system proposals, you’ll help shape the future of AI in the justice system.
Along the way, you’ll learn how to evaluate whether AI is the right fit for a task, design human–AI teams that work, build privacy-forward and trustworthy systems, and navigate the policy and change-management challenges of introducing AI into high-stakes environments.
By the end, your team will have produced a substantial, real-world deliverable—such as a UX research report, benchmark dataset, evaluation rubric, system design proposal, or prototype concept—giving you practical experience in public interest technology, AI system design, and leadership engagement. This is your chance to create AI that works for people, in practice, where it’s needed most.
The Stanford Legal Design Lab hosted its second annual AI & Access to Justice Summit as a gathering for leaders from legal aid organizations, technology companies, academia, philanthropists, and private practice. This diverse assembly of professionals gathered to discuss the potential of generative AI, and — most crucially at this moment of Autumn 2025 — to strategize about how to make AI work at scale to address the justice gap.
The summit’s mission was clear: to move beyond the hype cycle and forge a concrete path forward for a sustainable AI & A2J ecosystem across the US and beyond. The central question posed was how the legal community could work as an ecosystem to harness this technology, setting an agenda for 2, 5, and 10-year horizons to create applications, infrastructure, and new service/business models that can get more people access to justice.
The Arc of the Summit
The summit was structured over 2 days to help the diverse participants learn about AI tools, pilots, case studies, and lessons learned for legal teams — and then giving the participants the opportunity to design new interventions and strategies for a stronger AI R&D ecosystem.
Day 1 was dedicated to learning and inspiration, featuring a comprehensive slate of speakers who presented hands-on demonstrations of cutting-edge AI tools, shared detailed case studies of successful pilots, and offered insights from the front lines of legal tech innovation.
Day 1 -> Day 2’s mission
Day 2 was designed to shift the focus from listening to doing, challenging attendees to synthesize the previous day’s knowledge into strategic designs, collaborative agendas, and new partnerships. This structure was designed to build a shared foundation of knowledge before embarking on the collaborative work of building the future.
The Summit began by equipping attendees with a new arsenal of technological capabilities, showcasing the tools that serve as the building blocks for this new era in justice.
Our Key AI + A2J Ecosystem Moment
The key theme of this year’s AI+A2J Summit is building a strong, coordinated R&D ecosystem. This is because our community of legal help providers, researchers, public interest tech-builders, and strategists are at a key moment.
It’s been over 3 years now since the launch of ChatGPT. Where are we going with AI in access to justice?
We are several years into the LLM era now — past the first wave of surprise, demos, and hype — and into the phase where real institutions are deciding what to do with these tools. People are already using AI to solve problems in their everyday lives, including legal problems, whether courts and legal aid organizations are ready or not. That means the “AI moment” is no longer hypothetical: it’s shaping expectations, workflows, and trust right now. But still many justice leaders are confused, overwhelmed, or unsure about how to get to positive impact in this new AI era.
Leaders are not sure how to make progress.
This is exactly why an AI+A2J Summit like this matters. We’re at a pivot point where the field can either coordinate and build durable public-interest infrastructure — or fragment into disconnected experiments that don’t translate into meaningful service capacity. A2J leaders are balancing urgency with caution, and the choices made in the next year or two will set patterns that could last a decade: what gets adopted, what gets regulated, what gets trusted, and what gets abandoned.
What will 2030 look like for A2J?
We have possible rosy futures and we have more devastating ones.
Which of these possible near futures will we have in 2030 for access to justice?
A robust, accessible marketplace of services — where everyone having a problem with their landlord, debt collector, spouse, employer, neighbor, or government can easily get the help they need in the form they want?
Or will we have a hugely underserved public, that’s frustrated and angry, facing an ever-growing asymmetry of robo-filed lawsuits and relying on low-quality AI help?
What is stopping great innovation imapact?
Some of the key things that could stop our community from delivering great outcomes in the next five years include a few big trends:
too much chilling regulation,
under-performing and -safety tested solutions that lead to bad harms and headlines,
not enough money flowing to get to solutions, everyone reinventing the wheel on their own and deliverting fragile and costly local solutions, and
a lack of a building substantive, meaninful solutions — instead focusing on small, peripheral tasks.
The primary barriers are not just technical — they’re operational, institutional, and human. Legal organizations need tools that are reliable enough to use with real people, real deadlines, and real consequences. But today, many pilots struggle with consistency, integration into daily workflows, and the basic “plumbing” that makes technology usable at scale: identity management, knowledge management, access controls, and clear accountability when something goes wrong.
Trust is also fragile in high-stakes settings, and the cost of a failure is unusually high. A single under-tested tool can create public harm, undermine confidence internally, and trigger an overcorrection that chills innovation. In parallel, many organizations are already stretched thin and running on complex legacy systems. Without shared standards, shared evaluation, and shared implementation support, the burden of “doing AI responsibly” becomes too heavy for individual teams to carry alone.
At the Summit, we worked on 3 different strategy levels to try to prevent these blocks from pushing us to low impact or a continued status quo.
3 Levels of Strategic Work to Set us towards a Good Ecosystem
The goal of the Summit was to get leaders from across the A2J world to clearly define 3 levels of strategy. That means going beyond the usual strategic track — which is just defining the policies and tech agenda for their internal organization.
This meant focusing on both project mode (what are cool ideas and use cases) and also strategy mode — so we can shape where this goes, rather than react to whatever the market and technology delivers. We’re convening people who are already experimenting with AI in courts, legal aid, libraries, and community justice organizations, and we’re asking them to step back and make intentional choices about what they will build, buy, govern, and measure over the next 12–24 months. The point is to move from isolated pilots to durable capacity: tools that can be trusted, maintained, and integrated into real workflows, with clear guardrails for privacy, security, and quality.
To do that, the Summit is designed to push work at three linked levels of strategy.
The 3 levels of straegy
Strategy Level 1: Internal Org Strategy around AI
First is internal, organizational strategy: what each institution needs to do internally — data governance, procurement standards, evaluation protocols, staff training, change management, and the operational “plumbing” that makes AI usable and safe.
Strategy 2: Ecosystem Strategy
Second is ecosystem strategy, that covers how different A2J organizations can collaborate to increase capacity and impact.
Thinking through an Ecosystem approach to share capacity and improve outcomes
This can scope out what we should build together — shared playbooks, common evaluation and certification approaches, interoperable data and knowledge standards, and shared infrastructure that prevents every jurisdiction from reinventing fragile, costly solutions.
Strategy 3: Towards Big Tech & A2J
Third is strategy vis-à-vis big tech: how the justice community can engage major AI platform providers with clear expectations and leverage — so the next wave of product decisions, safety defaults, partnerships, and pricing structures actually support access to justice rather than widen gaps.
As more people and providers go to Big Tech for their answers and development work, how do we get to better A2J impact and outcomes?
The Summit is ultimately about making a coordinated, public-interest plan now — so that by 2030 we have a legal help ecosystem that is more trustworthy, more usable, more interoperable, and able to serve far more people with far less friction.
The Modern A2J Toolbox: A Growing set of AI-Powered Solutions
Equipping justice professionals with the right technology is a cornerstone of modernizing access to justice. The Summit provided a tour of AI tools available to the community, ranging from comprehensive legal platforms designed for large-scale litigation to custom-built solutions tailored for specific legal aid workflows. This tour of the growing AI toolbox revealed an expanding arsenal of capabilities designed to augment legal work, streamline processes, and extend the reach of legal services.
Research & Case Management Assistants
Teams from many different AI and legal tech teams presented their solutions and explained how they can be used to expand access to justice.
Notebook LM: The Notebook LM tool from Google empowers users to create intelligent digital notebooks from their case files and documents. Its capabilities have been significantly enhanced, featuring an expanded context window of up to 1 million tokens, allowing it to digest and analyze vast amounts of information. The platform is fully multilingual, supporting over 100 languages for both queries and content generation. This enables it to generate a wide range of work products, from infographics and slide decks to narrated video overviews, making it a versatile tool for both internal analysis and client communication.
Harvey:Harvey is an AI platform built specifically for legal professionals, structured around three core components. The Assistant functions as a conversational interface for asking complex legal questions based on uploaded files and integrated research sources like LexisNexis. The Vault serves as a secure repository for case documents, enabling deep analysis across up to 10,000 different documents at once. Finally, Workflows provide one-click solutions for common, repeatable tasks like building case timelines or translating documents, with the ability for organizations to create and embed their own custom playbooks.
Thomson Reuters’ CoCounsel: CoCounsel is designed to leverage an organization’s complete universe of information — from its own internal data and knowledge management systems to the primary law available through Westlaw. This comprehensive integration allows it to automate and assist with tasks across the entire client representation lifecycle, from initial intake and case assessment to legal research and discovery preparation. The platform is built to function like a human colleague, capable of pulling together disparate information sources to efficiently construct the building blocks of legal practice. TR also has an AI for Justice program that leverages CoCounsel and its team to help legal aid organizations.
VLex’s Vincent AI:Vincent AI adopts a workflow-based approach to legal tasks, offering dedicated modules for legal research, contract analysis, complaint review, and large-scale document review. Its design is particularly user-friendly for those with “prompting anxiety,” as it can automatically analyze an uploaded document (such as a lease or complaint) and suggest relevant next steps and analyses. A key feature is its ability to process not just text but also audio and video content, opening up powerful applications for tasks like analyzing client intake calls or video interviews to rapidly identify key issues.
AI on Case Management & E-Discovery Platforms
Legal Server: As a long-standing case management system, Legal Server has introduced an AI assistant named “Ellis.” The platform’s core approach to AI is rooted in data privacy and relevance. Rather than drawing on the open internet, Ellis is trained exclusively on an individual client organization’s own isolated data repository, including its help documentation, case notes, and internal documents. This ensures that answers are grounded in the organization’s specific context and expertise while maintaining strict client confidentiality.
Relativity:Relativity’s e-discovery platform is made available to justice-focused organizations through its “Justice for Change” program. The platform includes powerful generative AI features like AIR for Review, which can analyze hundreds of thousands of documents to identify key people, terms, and events in an investigation. It also features integrated translation tools that support over 100 languages, including right-to-left languages like Hebrew, allowing legal teams to seamlessly work with multilingual case documents within a single, secure environment.
These tools represent a leap in technological capability. They all show the growing ability for AI to help legal teams synthesize info, work with documents, conduct research, produce key work product, and automate workflows. But how do we go from tech tools to real-world impact, solutions that are deployed at scale and get to high performance numbers? The Summit moved from tech demos to case studies to get to accounts of how to get to value and impact.
From Pilots to Impact: AI in Action Across the Justice Sector
In the second half of Day 1, the Summit moved beyond product demonstrations to showcase a series of compelling case studies from across the justice sector. These presentations offered proof points of how organizations are already leveraging AI to serve more people, improve service quality, and create new efficiencies, delivering concrete value to their clients and communities today.
Legal Aid Society of Middle Tennessee & The Cumberlands — Automating Expungement Petitions:The “ExpungeMate” project was created to tackle the manual, time-consuming process of reviewing criminal records and preparing expungement petitions. By building a custom GPT to analyze records and an automated workflow to generate the necessary legal forms, the organization dramatically transformed its expungement clinics. At a single event, their output surged from 70 expungements to 751. This newfound efficiency freed up attorneys to provide holistic advice and enabled a more comprehensive service model that brought judges, district attorneys, and clerks on-site to reinstate driver’s licenses and waive court debt in real-time.
Citizens Advice (UK) — Empowering Advisors with Caddy: Citizens Advice developed Caddy (Citizens Advice Digital Assistant), an internal chatbot designed to support its network of advisors, particularly new trainees. Caddy uses a Retrieval-Augmented Generation (RAG), a method that grounds the AI’s answers in a private, trusted knowledge base to ensure accuracy and prevent hallucination. A key feature is its “human-in-the-loop” workflow, where supervisors can quickly validate answers before they are given to clients. A six-week trial demonstrated significant impact, with the evaluation found that Caddy halved the response time for advisors seeking supervisory support, unlocking capacity to help thousands more people.
Frontline Justice — Supercharging Community Justice Workers To support its network of non-lawyer “justice workers” in Alaska, Frontline Justice deployed an AI tool designed not just as a Q&A bot, but as a peer-to-peer knowledge hub. While the AI provides initial, reliable answers to legal questions, the system empowers senior justice workers to review, edit, and enrich these answers with practical, on-the-ground knowledge like local phone numbers or helpful infographics. This creates a dynamic, collaborative knowledge base where the expertise of one experienced worker in a remote village can be instantly shared with over 200 volunteers across the state.
Lone Star Legal Aid — Building a Secure Chatbot Ecosystem Lone Star Legal Aid embarked on an ambitious in-house project to build three distinct chatbots on a secure RAG architecture to serve different user groups. One internal bot, LSLAsks, is for administrative information in their legal aid group. Their internal bot for legal staff, Juris, was designed to centralize legal knowledge and defeat the administrative burden of research. A core part of their strategy involved rigorous A/B testing of four different search models (cleverly named after the Ninja Turtles) to meticulously measure accuracy, relevancy, and speed, with the ultimate goal of eliminating hallucinations and building user trust in the system.
People’s Law School (British Columbia) — Ensuring Quality in Public-Facing AI The team behind the public-facing Beagle+ chatbot shared their journey of ensuring high-quality, reliable answers for the public. Their development process involved intensive pre- and post-launch evaluation. Before launch, they used a 42-question dataset of real-world legal questions to test different models and prompts until they achieved 99% accuracy. After launch, a team of lawyers reviewed every single one of the first 5,400 conversations to score them for safety and value, using the findings to continuously refine the system and maintain its high standard of quality.
These successful implementations offered more than just inspiration; they surfaced a series of critical strategic debates that the entire access to justice community must now navigate.
Lessons Learned and Practical Strategies from the First Generation of AI+A2J Work
A consistent “lesson learned” from Day 1 was that legal aid AI only works when it’s treated as mission infrastructure, not as a cool add-on. Leaders emphasized values as practical guardrails: put people first (staff + clients), keep the main thing the main thing (serving clients), and plan for the long term — especially because large legal aid organizations are “big ships” that can’t pivot overnight.
Smart choice of projects: In practice, that means choosing projects that reduce friction in frontline work, don’t distract from service delivery, and can be sustained after the initial burst of experimentation.
An ecosystem of specific solutions: On the build side, teams stressed scoping and architecture choices that intentionally reduce risk. One practical pattern was a “one tool = one problem” approach, with different bots for different users and workflows (internal legal research, internal admin FAQs, and client-facing triage) rather than trying to make a single chatbot do everything.
Building for Security & Privacy forward solutions: Security and privacy were treated as design requirements, not compliance afterthoughts — e.g., selecting an enterprise cloud environment already inside the organization’s security perimeter and choosing retrieval-augmented generation (RAG) to keep answers grounded in verified sources.
Keeping Knowledge Fresh: Teams also described curating the knowledge base (black-letter law + SME guidance) and setting a maintenance cadence so the sources stay trustworthy over time.
Figure out What You’re Measuring & How: On evaluation, Day 1 emphasized that “accuracy” isn’t a vibe — you have to measure it, iterate, and keep monitoring after launch. Practical approaches included: (1) building a small but meaningful test set from real questions, (2) defining what an “ideal answer” must include, and (3) scoring outputs on safety and value across model/prompt/RAG variations.
Teams also used internal testing with non-developer legal staff to ask real workflow questions, paired with lightweight feedback mechanisms (thumbs up/down + reason codes) and operational metrics like citations used, speed, and cost per question. A key implementation insight was that some “AI errors” are actually content errors — post-launch quality improved by fixing source content (even single missing words) and tightening prompts, supported by ongoing monitoring.
Be Ready with Policies & Governance: On deployment governance, teams highlighted a bias toward containment, transparency, and safe failure modes. One practical RAG pattern: show citations down to the page/section, display the excerpt used, and if the system can’t answer from the verified corpus, it should say so — explicitly.
There was also a clear warning about emerging security risks (especially prompt injection and attack surfaces when tools start browsing or pulling from the open internet) and the need to think about cybersecurity as capability scales from pilots to broader use. Teams described practical access controls (like 2FA) and “shareable internal agents” as ways to grow adoption without losing governance.
Be Ready for Data Access Blocks: Several Day 1 discussions surfaced the external blockers that legal aid teams can’t solve alone — especially data access and interoperability with courts and other systems.
Even when internal workflows are ready, teams run into constraints like restrictions on scraping or fragmented, jurisdiction-specific data practices, which makes replication harder and increases costs for every new deployment. That’s one reason the “lessons learned” kept circling back to shared infrastructure: common patterns for grounded knowledge, testing protocols, security hardening, and the data pathways needed to make these tools reliable in day-to-day legal work.
Strategic Crossroads: Key Debates Shaping the Future of the AI+A2J Ecosystem
The proliferation of AI has brought the access to justice community to a strategic crossroads. The Summit revealed that organizations are grappling with fundamental decisions about how to acquire, build, and deploy this technology. The choices made in the coming years will define the technological landscape of the sector, determining the cost, accessibility, and control that legal aid organizations have over their digital futures.
The Build vs. Buy Dilemma
A central tension emerged between building custom solutions and purchasing sophisticated off-the-shelf platforms. We might end up with a ‘yes and’ approach, that involves both.
The Case for Building:
Organizations like Maryland Legal Aid and Lone Star Legal Aid are pursuing an in-house development path. This is not just a cost-and-security decision but a strategic choice about building organizational capacity.
The primary drivers are significantly lower long-term costs — Maryland Legal Aid reported running their custom platform for their entire staff for less than $100 per month — and enhanced data security and privacy, achieved through direct control over the tech stack and zero-data-retention agreements with API providers.
Building allows for the precise tailoring of tools to unique organizational workflows and empowers staff to become creators.
The Case for Buying:
Conversely, presentations from Relativity, Harvey, Thomson Reuters, vLex/Clio, and others showcased the immense power of professionally developed, pre-built platforms. The argument for buying centers on leveraging cutting-edge technology and complex features without the significant upfront investment in hiring and maintaining an in-house development team.
This path offers immediate access to powerful tools for organizations that lack the capacity or desire to become software developers themselves.
Centralized Expertise vs. Empowered End-Users
A parallel debate surfaced around who should be building AI applications. The traditional model, exemplified by Lone Star Legal Aid, involves a specialized technical team that designs and develops tools for the rest of the organization.
In contrast, Maryland Legal Aid presented a more democratized vision, empowering tech-curious attorneys and paralegals to engage in “vibe coding.”
This approach envisions non-technical staff becoming software creators themselves, using new, user-friendly AI development tools to rapidly build and deploy solutions. It transforms end-users into innovators, allowing legal aid organizations to “start solving their own problems” fast, cheaply, and in-house.
Navigating the Role of Big Tech in Justice Services
The summit highlighted the inescapable and growing role of major technology companies in the justice space. The debate here centers on the nature of the engagement.
One path involves close collaboration, such as licensing tools like Notebook LM from Google or leveraging APIs from OpenAI to power custom applications.
The alternative is a more cautious approach that prioritizes advocacy for regulation, taxation and licensing legal orgs’ knowledge and tools, and the implementation of robust public interest protections to ensure that the deployment of large-scale AI serves, rather than harms, the public good.
These strategic debates are shaping the immediate future of legal technology, but the summit also issued a more profound challenge: to use this moment not just to optimize existing processes, but to reimagine the very foundations of justice itself.
AI Beyond Automation: Reimagining the Fundamentals of the Justice System
The conversation at the summit elevated from simply making the existing justice system more efficient to fundamentally transforming it for a new era.
In a thought-provoking remote address, Professor Richard Susskind challenged attendees to look beyond the immediate applications of AI and consider how it could reshape the core principles of dispute resolution and legal help. This forward-looking perspective urged the community to avoid merely automating the past and instead use technology to design a more accessible, preventative, and outcome-focused system of justice.
The Automation Fallacy
Susskind warned against what he termed “technological myopia” — the tendency to view new technology only through the lens of automating existing tasks. He argued that simply replacing human lawyers with AI to perform the same work is an uninspired goal. Using a powerful analogy, he urged the legal community to avoid focusing on the equivalent of “robotic surgery” (perfecting an old process) and instead seek out the legal equivalents of “non-invasive therapy” and “preventative medicine” — entirely new, more effective ways to achieve just outcomes.
Focusing Upstream
This call to action was echoed in a broader directive to shift focus from downstream dispute resolution to upstream interventions. The goal is to leverage technology and data not just to manage conflicts once they arise, but to prevent them from escalating in the first place. This concept was vividly captured by Susskind’s metaphor of a society that is better served by “putting a fence at the top of the cliff rather than an ambulance at the bottom.”
The Future of Dispute Resolution
Susskind posed the provocative question, “Can AI replace judges?” but quickly reframed it to be more productive. Instead of asking if a machine can replicate a human judge, he argued the focus should be on outcomes: can AI systems generate reliable legal determinations with reasons?
He envisioned a future, perhaps by 2030, where citizens might prefer state-supported, AI-underpinned dispute services over traditional courts. In this vision, parties could submit their evidence and arguments to a “comfortingly branded” AI system that could cheaply, cheerfully, and immediately deliver a conclusion, transforming the speed and accessibility of justice.
Achieving such ambitious, long-term visions requires more than just technological breakthroughs; it demands the creation of a practical, collaborative infrastructure to build and sustain this new future.
Building Funding and Capacity for this Work
On the panel about building a National AI + A2J ecosystem, panelists discussed how to increase capacity and impact in this space.
The Need to Make this Space Legible as a Market
The panel framed the “economics” conversation as a market-making challenge: if we want new tech to actually scale in access to justice, we have to make the space legible — not just inspiring. There could be a clearer market for navigation tech in low-income “fork-in-the-road” moments. The panel highlighted that the nascent ecosystem needs three things to become investable and durable:
clearly defined problems,
shared infrastructure that makes building and scaling easier, and
business models that sustain products over time.
A key through-line in the panel’s commentary was: we can’t pretend grant funding alone will carry the next decade of AI+A2J delivery. Panelists suggested we need experimentation to find new payers — for example, employer-funded benefits and EAP dollars, or insurer/health-adjacent funding tied to social determinants of health — paired with stronger evidence that tools improve outcomes. This is connected to the need for shared benchmarks and evaluation methods that can influence how developers build and how funders (and institutions) decide what to back.
A Warning Not to Build New Tech on Bad Processes
The panel also brought a grounding reality check: even the best tech will underperform — or do harm — if it’s layered onto broken processes. Tech projects where tech sat on top of high-default systems contributed to worse outcomes.
The economic implication was clear: funders and institutions should pay for process repair and procedural barrier removal as seriously as they pay for new tools, because the ROI of AI depends on the underlying system actually functioning.
The Role of Impact Investing as a new source of capital
Building this ecosystem requires a new approach to funding. Kate Fazio framed the justice gap as a fundamental “market failure” in the realm of “people law” — the everyday legal problems faced by individuals. She argued that the two traditional sources of capital are insufficient to solve this failure: traditional venture capital is misaligned, seeking massive returns that “people law” cannot generate, while philanthropy is vital but chronically resource-constrained.
The missing piece, Fazio argued, isimpact investing: a form of patient, flexible capital that seeks to generate both a measurable social impact and a financial return. This provides a crucial middle ground for funding sustainable, scalable models that may not offer explosive growth but can create enormous social value. But she highlighted a stark reality: of the 17 UN Sustainable Development Goals, Goal 16 (Peace, Justice, and Strong Institutions) currently receives almost no impact investment capital. This presents both a monumental challenge and a massive opportunity for the A2J community to articulate its value and attract a new, powerful source of funding to build the future of justice.
This talk of new capital, market-making, and funding strategies started to point the group to a clear strategic imperative. To overcome the risk of fragmented pilots and siloed innovation, the A2J community must start coalescing into a coherent ecosystem. This means embracing collaborative infrastructure, which can be hand-in-hand with attracting new forms of capital.
By reframing the “market failure” in people law as a generational opportunity for impact investing, the sector can secure the sustainable funding needed to scale the transformative, preventative, and outcome-focused systems of justice envisioned throughout the summit.
Forging an AI+A2J Ecosystem: The Path to Sustainable Scale and Impact
On Day 2, we challenged groups to envision how to build a strong AI and A2J development, evaluation, and market ecosystem. They came up with so many ideas, and we try to capture them below. Much of it is about having common infrastructure, shared capacity, and better ways to strengthen and share organic DIY AI tools.
A significant risk facing the A2J community is fragmentation, a scenario where “a thousand pilots bloom” but ultimately fail to create lasting, widespread change because efforts are siloed and unsustainable. The summit issued a clear call to counter this risk by adopting a collaborative ecosystem approach.
The working groups on Day 2 highlighted some of the key things that our community can work on, to build a stronger and more successful A2J provider ecosystem. This infrastructure-centered strategy emphasizes sharing knowledge, resources, and infrastructure to ensure that innovations are not only successful in isolation but can be sustained, scaled, and adapted across the entire sector.
Throughout the summit, presenters and participants highlighted the essential capacities and infrastructure that individual organizations must develop to succeed with AI. Building these capabilities in every single organization is inefficient and unrealistic. An ecosystem approach recognizes the need for shared infrastructure, including the playbooks, knowledge/data standards, privacy and security tooling, evaluation and certification, and more.
Replicable Playbooks to Prevent Parallel Duplication
Many groups in the Summit called for replicable solutions playbooks, that go beyond sharing repositories on Github and making conference presentations, and getting to the teams and resources that can help more legal teams replicate successful AI solutions and localize them to their jurisdiction and organization.
A2J organizations don’t just need inspiration — they need proven patterns they can adopt with confidence. Replicable “how-tos” turn isolated success stories into field-level capability: how to scope a use case, how to choose a model approach, how to design a safe workflow, how to test and monitor performance, and how to roll out tools to staff without creating chaos. These playbooks reduce the cost of learning, lower risk, and help organizations move from pilots to sustained operations.
Replicable guidance also helps prevent duplication. Right now, too many teams are solving the same early-stage problems in parallel: procurement questions, privacy questions, evaluation questions, prompt and retrieval design, and governance questions. If the field can agree on shared building blocks and publish them in usable formats, innovation becomes cumulative — each new project building on the last instead of starting over.
A Common Agenda of What Tasks-Issues to Build Solutions for
Without a shared agenda, the field risks drifting into fragmentation: dozens of pilots, dozens of platforms, and no cumulative progress. A common agenda does not mean one centralized solution — it means alignment on what must be built together, what must be measured, and what must be stewarded over time. It creates shared language, shared priorities, and shared accountability across courts, legal aid, community organizations, researchers, funders, and vendors.
This is the core reason the Legal Design Lab held the Summit: to convene the people who can shape that shared agenda and to produce a practical roadmap that others can adopt. The goal is to protect this moment from predictable failure modes — over-chill, backlash, duplication, and under-maintained tools — and instead create an ecosystem where responsible innovation compounds, trust grows, and more people get real legal help when they need it.
Evaluation Protocols and Certifications
Groups also called for more, easier evaluation and certification. They want high-quality, standardized methods for evaluation, testing, and long-term maintenance.
In high-stakes legal settings, “seems good” is not good enough. The field needs clear definitions of quality and safety, and credible evaluation protocols that different organizations can use consistently. This doesn’t mean one rigid standard for every tool — but it does mean shared expectations: what must be tested, what must be logged, what harms must be monitored, and what “good enough” looks like for different risk levels.
Certification — or at least standard conformance levels — can also shift the market. If courts and legal aid groups can point to transparent evaluation and safety practices, then vendors and internal builders alike have a clear target. That reduces fear-driven overreaction and replaces it with evidence-driven decision-making. Over time, it supports responsible procurement, encourages better products, and protects the public by making safety and accountability visible.
In addition, creating legal benchmarksfor the most common & significant legal tasks can push LLM developers to improve their foundational models for justice use cases
Practical, Clear Privacy Protections
A block for many of the possible solutions is the safe use of AI with highly confidential, risky data. Privacy is not a footnote in A2J — it is the precondition for using AI with real people. Many of the highest-value workflows involve sensitive information: housing instability, family safety, immigration status, disability, finances, or criminal history. If legal teams cannot confidently protect client data, they will either avoid the tools entirely or use them in risky ways that expose clients and organizations to harm.
What is needed is privacy-by-design infrastructure: clear rules for data handling, retention, and access; secure deployment patterns; strong vendor contract terms; and practical training for staff about what can and cannot be used in which tools. The Summit is a place to align on what “acceptable privacy posture” should look like across the ecosystem — so privacy does not become an innovation-killer, and innovation does not become a privacy risk.
More cybersecurity, testing, reliability engineering, and ongoing monitoring
Along with privacy risks, participants noted that many of the organic, DIY solutions are not prepared for cybersecurity risks. As AI tools become embedded in legal workflows, they become targets — both for accidental failures and deliberate attacks. Prompt injection, data leakage, insecure integrations, and overbroad permissions can turn a helpful tool into a security incident. And reliability matters just as much as brilliance: a tool that works 80% of the time may still be unusable in high-stakes practice if the failures are unpredictable.
The field needs a stronger norm of “safety engineering”: threat modeling, red-teaming, testing protocols, incident response plans, and ongoing monitoring after deployment. This is also where shared infrastructure helps most. Individual organizations should not each have to invent cybersecurity practices for AI from scratch. A common set of testing and security baselines would let innovators move faster while reducing systemic risk.
Inter-Agency/Court Data Connections
Many groups need to call up and work with data from other agencies — like court docket files and records, other legal aid groups’ data, and more — in order to get highly effective, AI-powered workflows
Participants called for more standards and data contracts that can facilitate systematic data access, collection, and preparation. Many of the biggest A2J bottlenecks are not about “knowing the law” — they’re about navigating fragmented systems. People have to repeat their story across multiple offices, programs, and portals. Providers can’t see what happened earlier in the journey. Courts don’t receive information in consistent, structured ways. The result is duplication, delay, and drop-off — exactly where AI could help, but only if the data ecosystem supports it.
Many of the biggest A2J bottlenecks are not about “knowing the law” — they’re about navigating fragmented systems. People have to repeat their story across multiple offices, programs, and portals. Providers can’t see what happened earlier in the journey. Courts don’t receive information in consistent, structured ways. The result is duplication, delay, and drop-off — exactly where AI could help, but only if the data ecosystem supports it.
Data Contracts for Interoperable Knowledge Bases
Many local innovators are starting to build out structured, authoritative knowledge on court procedure, forms and documents, strategies, legal authorities, service directories, and more. This knowledge data is built to power their local legal AI solutions, but right now it is stored and saved in unique local ways.
This investment in local authoritative legal knowledge bases makes sense. LLMs are powerful, but they are not a substitute for authoritative, maintainable legal knowledge. The most dependable AI systems in legal help will be grounded in structured knowledge: jurisdiction-specific procedures, deadlines, forms, filing rules, court locations, service directories, eligibility rules, and “what happens next” pathways.
But the worry among participants is that all of these highly localized knowledge bases will be one-off for a specific org or solution. Ideally, when teams are investing in building these local knowledge bases, it can follow some key standard rules so it can perform well and it can be updated, audited, and reused across tools and regions.
This is why knowledge bases and data exchanges are central to the ecosystem approach. Instead of each organization maintaining its own isolated universe of content, we can build shared registries and common schemas that allow local control while enabling cross-jurisdiction learning and reuse. The aim is not uniformity for its own sake — it’s reliability, maintainability, and the ability to scale help without scaling confusion.
More training and change management so legal teams are ready
Even the best tools fail if people don’t adopt them in real workflows. Legal organizations are human systems with deeply embedded habits, risk cultures, and informal processes. Training and change management are not “nice to have” — they determine whether AI becomes a daily capability or a novelty used by a handful of early adopters.
What’s needed is practical, role-based readiness support: training for leadership on governance and procurement, training for frontline staff on safe use and workflow integration, and support for managers who must redesign processes and measure outcomes. The Summit is a step toward building a shared approach to readiness — so the field can absorb change without burnout, fragmentation, or loss of trust.
Building Capability & Lowering Costs of Development/Use
One of the biggest barriers to AI-A2J impact is that the “real” version of these tools — secure deployments, quality evaluation, integration into existing systems, and sustained maintenance — can be unaffordable when each court or legal aid organization tries to do it alone. The result is a familiar pattern: a few well-resourced organizations build impressive pilots, while most teams remain stuck with limited access, short-term experiments, or tools that can’t safely touch real client data.
Coordination is the way out of this trap. When the field aligns on shared priorities and shared building blocks, we reduce duplication and shift spending away from reinventing the same foundational components toward improving what actually matters for service delivery.
Through coordination, the ecosystem can also change the economics of AI itself. Shared evaluation protocols, reference architectures, and standard data contracts mean vendors and platform providers can build once and serve many — lowering per-organization cost and making procurement less risky. Collective demand can also create better terms: pooled negotiation for pricing, clearer requirements for privacy/security, and shared expectations about model behavior and transparency.
Just as importantly, coordinated open infrastructure — structured knowledge bases, service directories, and interoperable intake/referral data — reduces reliance on expensive bespoke systems by making high-value components reusable across jurisdictions.
The goal is not uniformity, but a commons: a set of shared standards and assets that makes safe, high-quality AI deployment feasible for the median organization, not just the best-funded one.
Conclusion
The AI + Access to Justice Summit is designed as a yearly convening point — because this work can’t be finished in a single event. Each year, we’ll take stock of what’s changing in the technology, what’s working on the ground in courts and legal aid, and where the biggest gaps remain. More importantly, we’ll use the Summit to move from discussion to shared commitments: clearer priorities, stronger relationships across the ecosystem, and concrete next steps that participants can carry back into their organizations and collaborations.
We are also building the Summit as a launchpad for follow-through. In the months after convening, we will work with participants to continue progress on common infrastructure: evaluation and safety protocols, privacy and security patterns, interoperable knowledge and data standards, and practical implementation playbooks that make adoption feasible across diverse jurisdictions. The aim is to make innovation cumulative — so promising work does not remain isolated in a single pilot site, but becomes reusable and improvable across the field.
We are deeply grateful to the sponsors who made this convening possible, and to the speakers who shared lessons, hard-won insights, and real examples from the frontlines.
Most of all, thank you to the participants — justice leaders, technologists, researchers, funders, and community partners — who showed up ready to collaborate, challenge assumptions, and build something larger than any single organization can create alone. Your energy and seriousness are exactly what this moment demands, and we’re excited to keep working together toward a better 2030.
The Stanford Legal Design Lab hosted the second annual AI and Access to Justice Summit on November 20-21, 2025. Over 150 legal professionals, technologists, regulators, strategists, and funders came together to tackle one big question: how can we build a strong, sustainable national/international AI and Access to Justice Ecosystem?
We will be synthesizing all of the presentations, feedback, proposals and discussions into a report that lays out:
The current toolbox that legal help teams and users can be employing to accomplish key legal tasks like Q&A, triage and referrals, conducting intake interviews, drafting documents, doing legal research, reviewing draft documents, and more.
The strategies, practical steps, and methods with which to design, develop, evaluate, and maintain AI so that it is valuable, safe, and affordable.
Exemplary case studies of what AI solutions are being built, how they are being implemented in new service and business models, and how they might be scaled or replicated.
An agenda of how to encourage more coordination of AI technology, evaluation, and capability-building, so that successful solutions can be available to as many legal teams and users as possible — and have the largest positive impact on people’s housing, financial, family, and general stability.
Thank you to all of our speakers, participants, and sponsors!
In the access to justice world, we often talk about “the justice gap” as if it’s one massive, monolithic challenge. But if we want to truly serve the public, we need to be more precise. People encounter different kinds of legal problems, with different stakes, emotional dynamics, and system barriers. And those differences matter.
At the Legal Design Lab, we find it helpful to divide the access to justice landscape into three distinct types of problems. Each has its own logic — and each requires different approaches to research, design, technology, and intervention.
3 Types of Conflicts that we talk about when we talk about Access to Justice
1. David vs. Goliath Conflicts
This is the classic imbalance. An individual — low on time, legal knowledge, money, or support — faces off against a repeat player: a bank, a corporate landlord, a debt collector, or a government agency.
These Goliaths have teams of lawyers, streamlined filing systems, institutional knowledge, predictive data, and now increasingly, AI-powered legal automation and strategies. They can file thousands of cases a month — many of which go uncontested because people don’t understand the process, can’t afford help, or assume there’s no point trying.
This is the world of:
Eviction lawsuits from corporate landlords
Mass debt collection actions
Robo-filed claims, often incorrect but rarely challenged
The problem isn’t just unfairness — it’s non-participation. Most “Davids” default. They don’t get their day in court. And as AI makes robo-filing even faster and cheaper, we can expect the imbalance in knowledge, strategy, and participation may grow worse.
What Goliath vs. David Conflicts need
Designing for this space means understanding the imbalance and structuring tools to restore procedural fairness. That might mean:
Tools that help people respond before defaulting. These could be pre-filing defense tools that detect illegal filings or notice issues. It could also be tools that prepare people to negotiate from a stronger position — or empower them to respond before defaulting.
Systems that detect and challenge low-quality filings. It could also involve systems that flag repeat abusive behavior from institutional actors.
Interfaces that simplify legal documents into plain language. Simplified, visual tools to help people understand their rights and the process quickly.
Research into procedural justice and scalable human-AI support models
2. Person vs. Person Conflicts
This second type of case is different. Here, both parties are individuals, and neither has a lawyer.
In this world, both sides are unrepresented and lack institutional or procedural knowledge. There’s real conflict — often with emotional, financial, or relational stakes — but neither party knows how to navigate the system.
Think about emotionally charged, high-stakes cases of everyday life:
Family law disputes (custody, divorce, child support)
Mom-and-pop landlord-tenant disagreements
Small business vs. customer conflicts
Neighbor disputes and small claims lawsuits
Both people are often confused. They don’t know which forms to use, how to prepare for court, how to present evidence, or what will persuade a judge. They’re frustrated, emotional, and worried about losing something precious — time with their child, their home, their reputation. The conflict is real and felt deeply, but both sides are likely confused about the legal process.
Often, these conflicts escalate unnecessarily — not because the people are bad, but because the system offers them no support in finding resolution. And with the rise of generative AI, we must be cautious: if each person gets an AI assistant that just encourages them to “win” and “fight harder,” we could see a wave of escalation, polarization, and breakdowns in courtrooms and relationships.
We have to design for a future legal system that might, with AI usage increasing, become more adversarial, less just, and harder to resolve.
What Person Vs. Person Justice Conflicts Need
In person vs. person conflicts, the goal should be to get to mutual resolutions that avoid protracted ‘high’ conflict. The designs needed are about understanding and navigation, but also about de-escalation, emotional intelligence, and procedural scaffolding.
Tools that promote resolution and de-escalation, not just empowerment. They can ideally support shared understanding and finding a solution that can work for both parties.
Shared interfaces that help both parties prepare for court fairly. Technology can help parties prepare for court, but also explore off-ramps like mediation.
Mediation-oriented AI prompts and conflict-resolution scaffolding. New tools could have narrative builders that let people explain their story or make requests without hostility. AI prompts and assistants could calibrate to reduce conflict, not intensify it.
Design research that prioritizes relational harm and trauma awareness.
This is not just a legal problem. It’s a human problem — about communication, trust, and fairness. Interventions here also need to think about parties that are not directly involved in the conflict (like the children in a family law dispute between separating spouses).
3. Person vs. Bureaucracy
Finally, we have a third kind of justice issue — one that’s not so adversarial. Here, a person is simply trying to navigate a complex system to claim a right or access a service.
These kinds of conflicts might be:
Applying for public benefits, or appealing a denial
Dealing with a traffic ticket
Restoring a suspended driver’s license
Paying off fines or clearing a record
Filing taxes or appealing a tax decision
Correcting an error on a government file
Getting work authorization or housing assistance
There’s no opposing party. Just forms, deadlines, portals, and rules that seem designed to trip you up. People fall through the cracks because they don’t know what to do, can’t track all the requirements, or don’t have the documents ready. It’s not a courtroom battle. It’s a maze.
Here many of the people caught in these systems do have rights and options. They just don’t know it. Or they can’t get through all the procedural hoops to claim them. It’s a quiet form of injustice — made worse by fragmented service systems and hard-to-reach agencies.
What Person vs. Bureaucracy Conflicts Need
For people vs. bureaucracy conflicts, the key word is navigation. People need supportive, clarifying tools that coach and guide them through the process — and that might also make the process simpler to begin with.
Seamless navigation tools that walk people through every step. These could be digital co-pilots that walk people through complex government workflows, and keep them knowledgeable and encouraged at each step.
Clear eligibility screeners and document checklists. These could be intake simplification tools that flag whether the person is in the right place, and sets expectations about what forms someone needs and when.
Text-based reminders and deadline alerts, to keep people on top of complicated and lengthy processes. These procedural coaches can keep people from ending up in endless continuances or falling off the process altogether. Personal timelines and checklists can track each step and provide nudges.
Privacy-respecting data sharing so users don’t have to “start over” every time. This could mean administrative systems that have document collection & data verication systems that gather and store proofs (income, ID, residence) that people need to supply over and again. It could also mean bringing their choices and details among trusted systems, so they don’t need to fill in another form.
This space is ripe for good technology. But it also needs regulatory design and institutional tech improvements, so that systems become easier to plug into — and easier to fix. Aside from user-facing designs, we also need to work on standardizing forms, moving from form-dependencies to structured data, and improve the tech operations of the systems.
Why These Distinctions Matter
These three types of justice problems are different in form, in emotional tone, and in what people need to succeed. That means we need to study them differently, run stakeholder sessions differently, evaluate them with slightly different metrics, and employ different design patterns and principles.
Each of these problem types requires a different kind of solution and ideal outcome.
In David vs. Goliath, we need defense, protection, and fairness. We need to help reduce the massive imbalance in knowledge, capacity, and relationships, and ensure everyone can have their fair day in court.
In Person vs. Person, we need resolution, dignity, and de-escalation. We need to help people focus on mutually agreeable, sustainable resolutions to their problems with each other.
In Person vs. Bureaucracy, we need clarity, speed, and guided action. We must aim for seamless, navigable, efficient systems.
Each type of problem requires different work by researchers, designers, an policymakers. These include different kinds of:
User research methods, and ways to bring stakeholders together for collaborative design sessions
Product and service designs, and the patterns of tools, interfaces, and messages that will engage and serve users in this conflict.
Evaluation criteria, about what success looks like
AI safety guidelines, about how to prevent bias, capture, inaccuracies, and other possible harms. We can expect these 3 different conflicts changing as more AI usage appears among litigants, lawyers, and court systems.
If we blur these lines, we risk building one-size-fits-none tools.
How might the coming wave of AI in the legal system affect these 3 different kinds of Access to Justice problems?
Toward Smarter Justice Innovation
At the Legal Design Lab, we believe this three-type framework can help researchers, funders, courts, and technologists build smarter interventions — and avoid repeating old mistakes.
We can still learn across boundaries. For example:
How conflict resolution tools from family law might help in small business disputes
How navigational tools in benefits access could simplify court prep
How due process protections in eviction can inform other administrative hearings
But we also need to be honest: not every justice problem is built the same. And not every innovation should look the same.
By naming and studying these three zones of access to justice problems, we can better target our interventions, avoid unintended harm, and build systems that actually serve the people who need them most.
The Stanford Legal Design Lab is proud to announce a new initiative funded by the Gates Foundation that aims to bring the power of artificial intelligence (AI) into the hands of legal aid professionals. With this new project, we’re building and testing AI systems—what we’re calling “AI co-pilots”—to support legal aid attorneys and staff in two of the most urgent areas of civil justice: eviction defense and reentry debt mitigation.
This work continues our Lab’s mission to design and deploy innovative, human-centered solutions that expand access to justice, especially for those who face systemic barriers to legal support.
A Justice Gap That Demands Innovation
Across the United States, millions of people face high-stakes legal problems without any legal representation. Eviction cases and post-incarceration debt are two such areas, where legal complexity meets chronic underrepresentation—leading to outcomes that can reinforce poverty, destabilize families, and erode trust in the justice system.
Legal aid organizations are often the only line of defense for people navigating these challenges, but these nonprofits are severely under-resourced. These organizations are on the front lines of help, but often are stretched thin with staffing, tech, and resources.
The Project: Building AI Co-Pilots for Legal Aid Workflows
In collaboration with two outstanding legal aid partners—Legal Aid Foundation of Los Angeles (LAFLA) and Legal Aid Services of Oklahoma (LASO)—we are designing and piloting four AI co-pilot prototypes: two for eviction defense, and two for reentry debt mitigation.
These AI tools will be developed to assist legal aid professionals with tasks such as:
Screening and intake
Issue spotting and triage
Drafting legal documents
Preparing litigation strategies
Interpreting complex legal rules
Rather than replacing human judgment, these tools are meant to augment legal professionals’ work. The aim is to free up time for higher-value legal advocacy, enable legal teams to take on more clients, and help non-expert legal professionals assist in more specialized areas.
The goal is to use a deliberate, human-centered process to first identify low-risk, high-impact tasks for AI to do in legal teams’ workflows, and then to develop, test, pilot, and evaluate new AI solutions that can offer safe, meaningful improvements to legal service delivery & people’s social outcomes.
Why Eviction and Reentry Debt?
These two areas were chosen because of their widespread and devastating impacts on people’s housing, financial stability, and long-term well-being.
Eviction Defense
Over 3 million eviction lawsuits are filed each year in the U.S., with the vast majority of tenants going unrepresented. Without legal advocacy, many tenants are unaware of their rights or defenses. It’s also hard to fill in the many complicated legal documents required to participate in they system, protect one’s rights, and avoid a default judgment. This makes it difficult to negotiate with landlords, comply with court requirements, and protect one’s housing and money.
Evictions often happen in a matter of weeks, and with a confusing mix of local and state laws, it can be hard for even experienced attorneys to respond quickly. The AI co-pilots developed through this project will help legal aid staff navigate these rules and prepare more efficiently—so they can support more tenants, faster.
Reentry Debt
When people return home after incarceration, they often face legal financial obligations that can include court fines, restitution, supervision fees, and other penalties. This kind of debt can make it hard for a person to get to stability with housing, employment, driver’s licenses, and family.
According to the Brennan Center for Justice, over 10 million Americans owe more than $50 billion in reentry-related legal debt. Yet there are few tools to help people navigate, reduce, or resolve these obligations. By working with LASO, we aim to prototype tools that can help legal professionals advise clients on debt relief options, identify eligibility for fee waivers, and support court filings.
What Will the AI Co-Pilots Actually Do?
Each AI co-pilot will be designed for real use in legal aid organizations. They’ll be integrated into existing workflows and tailored to the needs of specific roles—like intake specialists, paralegals, or staff attorneys. Examples of potential functionality include:
Summarizing client narratives and flagging relevant legal issues
Filling in common forms and templates based on structured data
Recommending next steps based on jurisdictional rules and case data
Generating interview questions for follow-up conversations
Cross-referencing legal codes with case facts
The design process will be collaborative and iterative, involving continuous feedback from attorneys, advocates, and technologists. We will pilot and evaluate each tool rigorously to ensure its effectiveness, usability, and alignment with legal ethics.
Spreading the Impact
While the immediate goal is to support LAFLA and LASO, we are designing the project with national impact in mind. Our team plans to publish:
Open-source protocols and sample workflows
Evaluation reports and case studies
Responsible use guidelines for AI in legal aid
Collaboration pathways with legal tech vendors
This way, other legal aid organizations can replicate and adapt the tools to their own contexts—amplifying the reach of the project across the U.S.
“There’s a lot of curiosity in the legal aid field about AI—but very few live examples to learn from,” Hagan said. “We hope this project can be one of those examples, and help the field move toward thoughtful, responsible adoption.”
Responsible AI in Legal Services
At the Legal Design Lab, we know that AI is not a silver bullet. Tools must be designed thoughtfully, with attention to risks, biases, data privacy, and unintended consequences.
This project is part of our broader commitment to responsible AI development. That means:
Using human-centered design
Maintaining transparency in how tools work and make suggestions
Prioritizing data privacy and user control
Ensuring that tools do not replace human judgment in critical decisions
Our team will work closely with our legal aid partners, domain experts, and the communities served to ensure that these tools are safe, equitable, and truly helpful.
Looking Ahead
Over the next two years, we’ll be building, testing, and refining our AI co-pilots—and sharing what we learn along the way. We’ll also be connecting with national networks of eviction defense and reentry lawyers to explore broader deployment and partnerships.
If you’re interested in learning more, getting involved, or following along with project updates, sign up for our newsletter or follow the Lab on social media.
We’re grateful to the Gates Foundation for their support, and to our partners at LAFLA and LASO for their leadership, creativity, and deep dedication to the clients they serve.
Together, we hope to demonstrate how AI can be used responsibly to strengthen—not replace—the critical human work of legal aid.
On October 17 and 18, 2024 Stanford Legal Design Lab hosted the first-ever AI and Access to Justice Summit.
The Summit’s primary goal was to build strong relationships and a national, coordinated roadmap of how AI can responsibly be deployed and held accountable to close the justice gap.
AI + A2J Summit at Stanford Law School
Who was at the Summit?
Two law firm sponsors, K&L Gates and DLA Piper, supported the Summit through travel scholarships, program costs, and strategic guidance.
The main group of invitees were frontline legal help providers at legal aid groups, law help website teams, and the courts. We know they are key players in deciding what kinds of AI should and could be impactful for closing the justice gap. They’ll also be key partners in developing, piloting, and evaluating new AI solutions.
Key supporters and regional leaders from bar foundations, philanthropies, and pro bono groups were also invited. Their knowledge about funding, scaling, past initiatives, and spreading projects from one organization and region to others was key to the Summit.
Technology developers also came, both from big technology companies like Google and Microsoft and legal technology companies like Josef, Thomson Reuters, Briefpoint, and Paladin. Some of these groups already have AI tools for legal services, but not all of them have focused in on access to justice use cases.
In addition, we invited researchers who are also developing strategies for responsible, privacy-forward, efficient ways of developing specialized AI solutions that could help people in the justice sphere, and also learn from how AI is being deployed in parallel fields like in medicine or mental health.
Finally, we had participants who work in regulation and policy-making at state bars, to talk about policy, ethics, and balancing innovation with consumer protection. The ‘rules of the road’ about what kinds of AI can be built and deployed, and what standards they need to follow, are essential for clarity and predictability among developers.
What Happened at the Summit?
The Summit was a 2-day event, split intentionally into 5 sections:
Hands-On AI Training: Examples and Research to upskill legal professionals. There were demo’s, explainers, and strategies about what AI solutions are already in use or possible for legal services. Big tech, legal tech, and computer science researchers presented participants with hands-on, practical, detailed tour of AI tools, examples, and protocols that can be useful in developing new solutions to close the justice gap.
Big Vision: Margaret Hagan and Richard Susskind opened up the 2nd day with a challenge: where does the access to justice community want to be in 2030 when it comes to AI and the justice gap? How can individual organizations collaborate, build common infrastructure, and learn from each other to reach our big-picture goals?
AI+A2J as of 2024: In the morning of the second day, two panels presented on what is already happening in AI and Access to Justice — including an inventory of current pilots, demo’s of some early legal aid chatbots, regulators’ guidelines, and innovation sandboxes. This can help the group all understand the early-stage developments and policies.
Design & Development of New Initiatives. In the afternoon of the second day, we led breakout design workshops on specific use cases: housing law, immigration law, legal aid intake, and document preparation. The diverse stakeholders worked together using our AI Legal Design workbook to scope out a proposal for a new solution — whether that might mean building new technology or adapting off-the-shelf tech to the needs.
Support & Collaboration. In the final session, we heard from a panel who could talk through support: financial support, pro bono partnership support, technology company licensing and architecture support, and other ways to build more new interdisciplinary relationships that could unlock the talent, strategy, momentum, and finances necessary to make AI innovation happen. We also discussed support around evaluation so that there could be more data and more feeling of safety in deploying these new tools.
Takeaways from the Summit
The Summit built strong relationships & common understanding among technologists, providers, researchers, and supporters. Our hope is that we can run the Summit annually, to track progress in tackling the justice gap with AI and to observe what progress has been made, year-to-year. It is also to see the development of these relationships, collaborations, and scaling of impact.
In addition, some key points emerged from the training, panels, workshops, and down-time discussions.
Common Infrastructure for AI Development
Though many AI pilots are going to have be local to a specific organization in a specific region, the national (or international) justice community can be working on common resources that can serve as infrastructure to support AI for justice.
Common AI Trainings: Regional leaders, who are newly being hired by state bars and bar foundations to train and explore how AI can fit with legal services, should be working together to develop common training, common resources, and common best practices.
Project Repository: National organizations and networks should be thinking about a common repository of projects. This inventory could track what tech provider is being used, what benchmark is being used for evaluation, what AI model is being deployed, what date it was fine-tuned on, and if and how others could replicate it.
Rules of the Road Trainings. National organizations and local regulators could give more guidance to leadership like legal aid executive directors about what is allowed or not allowed, what is risky or safe, or other clarification that can help more leadership be brave and knowledgeable about how to deploy AI responsibly. When is an AI project sufficiently tested to be released to the public? How should the team be maintaining and tracking an AI project, to ensure it’s mitigating risk sufficiently?
Public Education. Technology companies, regulators, and frontline providers need to be talking more about how to make sure that the AI that is already out there (like ChatGPT, Gemini, and Claude) is reliable, has enough guardrails, and is consumer-safe. More research needs to be done on how to encourage strategic caution among the public, so they can use the AI safely and avoid user mistakes with it (like overreliance or misunderstanding).
Regulators<->Frontline Providers. More frontline legal help providers need to be in conversation with regulators (like bar associations, attorneys general, or other state/federal agencies) to talk about their perspective on if and how AI can be useful in closing the justice gap. Their perspective on risks, consumer harms, opportunities, and needs from regulators can ensure that rules are being set to maximize positive impact and minimize consumer harm & technology chilling.
Bar Foundation Collaboration. Statewide funders (especially bar foundations) can be talking to each other about their funding, scaling, and AI strategies. Well-resourced bar foundations can share how they are distributing money, what kinds of projects they’re incentivizing, how they are holding the projects accountable, and what local resources or protocols they could share with others.
AI for Justice Should be Going Upstream & Going Big
Richard Susskind charged the group with thinking big about AI for justice. His charges & insights inspired many of the participants throughout the Summit, particularly on two points.
Going Big. Susskind called on legal leaders and technologists not to do piecemeal AI innovation (which might well be the default pathway). Rather, he called on them to work in coordination across the country (if not the globe). The focus should be on reimagining how to use AI as a way to make a fundamental, beneficial shift in justice services. This means not just doing small optimizations or tweaks, but shifting the system to work better for users and providers.
Susskind charged us with thinking beyond augmentation to models of serving the public with their justice needs.
Going Upstream. He also charged us with going upstream, figuring out more early ways to spot and get help to people. This means not just adding AI into the current legal aid or court workflow — but developing new service offerings, data links, or community partnerships. Can we prevent more legal problems by using AI before a small problem spirals into a court case or large conflict?
After Susskind’s remarks, I focused in on coordination among legal actors across the country for AI development. Compared to the last 20 years of legal technology development, are there ways to be more coordinated, and also more focused on impact and accountability?
There might be strategic leaders in different regions of the US and in different issue areas (housing, immigration, debt, family, etc) that are spreading
best practices,
evaluation protocols and benchmarks,
licensing arrangements with technology companies
bridges with the technology companies
conversations with the regulators.
How can the Access to Justice community be more organized so that their voice can be heard as
the rules of the road are being defined?
technology companies are building and releasing models that the public is going to be using?
technology vendors decide if and how they are going to enter this market, and what their pricing and licensing are going to look like?
Ideally, legal aid groups, courts, and bars will be collaborating together to build AI models, agents, and evaluations that can get a significant number of people the legal help they need to resolve their problems — and to ensure that the general, popular AI tools are doing a good job at helping people with their legal problems.
Privacy Engineering & Confidentiality Concerns
One of the main barriers to AI R&D for justice is confidentiality. Legal aid and other help providers have a duty to keep their clients’ data confidential, which restricts their ability to use past data to train models or to use current data to execute tasks through AI. In practice, many legal leaders are nervous about any new technology that requires client data — -will it lead to data leaks, client harms, regulatory actions, bad press or other concerning outcomes?
Our technology developers and researchers had cutting-edge proposals for privacy-forward AI development, that could deal with some of these concerns around confidentiality. THough these privacy engineering strategies are foreign to many lawyers, the technologists broke them down into step-by-step explanations with examples, to help more legal professionals be able to think about data protection in a systematic, engineering way.
Synthetic Data. One of the privacy-forward strategies discussed was synthetic data. With this solution, a developer doesn’t use real, confidential data to train a system. Rather, they create a parallel but fictional set of data — like a doppelganger to the original client data. It’s structurally similar to confidential client data, but it contains no real people’s information. Synthetic data is a common strategy in healthcare technology, where there is a similar emphasis on patient confidentiality.
Neel Guha explained to the participants how synthetic data works, and how they might build a synthetic dataset that is free of identifiable data and does not violate ethical duties to confidentiality. He emphasized that the more legal aid and court groups can develop datasets that are share-able to researchers and the public, the more that researchers and technologists will be attracted to working on justice-tech challenges. More synthetic datasets will both be ethically safe & beneficial to collaboration, scaling, and innovation.
Federated Model Training. Another privacy/confidentiality strategy was Federated Model Training. Google DeepMind team presented on this strategy, taking examples from the health system.
When multiple hospitals all wanted to work on the same project: training an AI model to better spot tuberculosis or other issues on lung X-rays. Each hospital wanted to train the AI model on their existing X-ray data, but they did not want to let this confidential data to leave their servers and go to a centralized server. Sharing the data would break their confidentiality requirements.
So instead, the hospitals decided to go with a Federated Model training protocol. Here, an original, first version of the AI model was taken from the centralized server and then put on each of the hospital’s localized servers. The local version of the AI model would look at that hospital’s X-ray data and train the model on them. Then they would send the model back to the centralized server and accumulate all of the learnings and trainings to make a smart model in the center. The local hospital data was never shared.
In this way, legal aid groups or courts could explore making a centralized model while still keeping each of their confidential data sources on their private, secure servers. Individual case data and confidential data stay local on the local servers, and the smart collective model lives at a centralized place and gradually gets smarter. This technique can also work for training the model over time so that the model can continue to get smart as the information and data continue to grow.
Towards the Next Year of AI for Access to Justice
The Legal Design Lab team thanks all of our participants and sponsors for a tremendous event. We learned so much and built new relationships that we look forward to deepening with more collaborations & projects.
We were excited to hear frontline providers walk away with new ideas, concrete plans for how to borrow from others’ AI pilots, and an understanding of what might be feasible. We were also excited to see new pro bono and funding relationships develop, that can unlock more resources in this space.
Stay tuned as we continue our work on AI R&D, evaluation, and community-building in the access to justice community. We look forward to working towards closing the justice gap, through technology and otherwise!
The Legal Design Lab is proud to announce a new monthly online, public seminar on AI & Access to Justice: Research x Practice.
At this seminar, we’ll be bringing together leading academic researchers with practitioners and policymakers, who are all working on how to make the justice system more people-centered, innovative, and accessible through AI. Each seminar will feature a presentation from either an academic or practitioner who is working in this area & has been gathering data on what they’re learning. The presentations could be academic studies about user needs or the performance of technology, or less academic program evaluations or case studies from the field.
We look forward to building a community where researchers and practitioners in the justice space can make connections, build new collaborations, and advance the field of access to justice.
Sign up for the AI&A2J Research x Practice seminar, every first Friday of the month on Zoom.
At the April 2024 Stanford Codex FutureLaw Conference, our team at Legal Design Lab both presented the research findings about users’ and subject matter experts’ approaches to AI for legal help, and to lead a half-day interdisciplinary workshop on what future directions are possible in this space.
Many of the audience members in both sessions were technologists interested in the legal space, who are not necessarily familiar with the problems and opportunities for legal aid groups, courts, and people with civil legal problems. Our goal was to help them understand the “access to justice” space and spot opportunities to which their development & research work could relate.
Some of the ideas that emerged in our hands-on workshop included the following possible AI + A2J innovations:
AI to Scan Scary Legal Documents
Several groups identified that AI could help a person, who has received an intimidating legal document — a notice, a rap sheet, an immigration letter, a summons and complaint, a judge’s order, a discovery request, etc. AI could let them take a picture of the document, synthesize the information, present it back with a summary of what it’s about, what important action items are, and how to get started on dealing with it.
It could make this document interactive through FAQs, service referrals, or a chatbot that lets a person understand and respond to it. It could help people take action on these important but off-putting documents, rather than avoid them.
Using AI for Better Gatekeeping of Eviction Notices & Lawsuits
One group proposed that a future AI-powered system could screen possible eviction notices or lawsuit filings, to check if the landlord or property manager has fulfilled all obligations and m
Landlords must upload notices.
AI tools review the notice: is it valid? have they done all they can to comply with legal and policy requirements? is there any chance to promote cooperative dispute resolution at this early stage?
If the AI lives at the court clerk level, it might help court staff better detect errors, deficiencies, and other problems that better help them allocate limited human review.
AI to empower people without lawyers to respond to a lawsuit
In addition, AI could help the respondent (tenant) prepare their side, helping them to present evidence, prep court documents, understand court hearing expectations, and draft letters or forms to send.
Future AI tools could help them understand their case, make decisions, and get work product created with little burden.
With a topic like child support modification, AI could help a person negotiate a resolution with the other party, or do a trial run to see how a possible negotiation might go. It could also change their tone, to take a highly emotional negotiation request and transform it to be more likely to get a positive, cooperative reply from the other party.
AI to make Legal Help Info More Accessible
Another group proposed that AI could be integrated into legal aid, law library, and court help centers to:
Better create and maintain inter-organization referrals, so there are warm handoffs and not confusing roundabouts when people seek help
Clearer, better maintained, more organized websites for a jurisdiction, with the best quality resources curated and staged for easy navigation
Multi-modal presentations, to make information available in different visual presentations and languages
Providing more information in speech-to-text format, conversational chats, and across different dialects. This was especially highlighted in immigration legal services.
AI to upskill students & pro bono clinics
Several groups talked about AI for training and providing expert guidance to staff, law students, and pro bono volunteers to improve their capacity to serve members of the public.
AI tools could be used in simulations to better educate people in a new legal practice area, and also supplement their knowledge when providing services. Expert practitioners can supply knowledge to the tools, that can then be used by novice practitioners so that they can provide higher-quality services more efficiently in pro bono or law student clinics.
AI could also be used in community centers or other places where community justice workers operate, to get higher quality legal help to people who don’t have access to lawyers or who do not want to use lawyers.
AI to improve legal aid lawyers’ capacity
Several groups proposed AI that could be used behind-the-scenes by expert legal aid or court help lawyers. They could use AI to automate, draft, or speed up the work that they’re already doing. This could include:
Improving intake, screening, routing, and summaries of possible incoming cases
Drafting first versions of briefs, forms, affidavits, requests, motions, and other legal writing
Documenting their entire workflow & finding where AI can fit in.
Cross-Cutting action items for AI+ A2J
Across the many conversations, some common tasks emerged that cross different stakeholders and topics.
Reliable AI Benchmarks:
We as a justice community need to establish solid benchmarks to test AI effectiveness. We can use these benchmarks to focus on relevant metrics.
In addition, we need to regularly report on and track AI performance at different A2J tasks.
This can help us create feedback loops for continuous improvement.
Data Handling and Feedback:
The community needs reliable strategies and rules for how to do AI work that respects obligations for confidentiality and privacy.
Can there be more synthetic datasets that still represent what’s happening in legal aid and court practice, so they don’t need to share actual client information to train models?
Can there be better Personally Identifiable Information (PII) redaction for data sharing?
Who can offer guidance on what kinds of data practices are ethical and responsible?
Low-Code AI Systems:
The justice community is never going to have large tech, data, or AI working groups within their legal aid or court organization. They are going to need low-code solutions that will let them deploy AI systems, fine-tune them, and maintain them without a huge technical requirement.
Overall, the presentation, Q&A, and workshop all pointed to enthusiasm for responsible innovation in the AI+A2J space. Tech developers, legal experts, and strategists are excited about the opportunity to improve access to justice through AI-driven solutions, and enhance efficiency and effectiveness in legal aid. With more brainstormed ideas for solutions in this space, now it is time to move towards R&D incubation that can help us understand what is feasible and valuable in practice.
In December 2023, our lab hosted a half-day workshop on AI for Legal Help.
Our policy lab class of law students, master students, and undergraduates presented their user research findings from their September through December research.
Our guests, including those from technology companies, universities, state bars, legal aid groups, community-based organizations, and advocacy/think takes, all worked together in break-out sessions to tackle some of the big policy and legal opportunities around AI in the space.
We thank our main class partners, the Technology Initiative Grant team from the Legal Services Corporation, for assisting us with the direction and main feedback to our class user research work.
Our organizing committee was pleased to receive many excellent submissions for the AI & A2J Workshop at Jurix on December 18, 2023. We were able to select half of the submissions for acceptance, and we extended the half-day workshop to be a full-day workshop to accommodate the number of submissions.
We are pleased to announce our final schedule for the workshop:
Schedule for the AI & A2J Workshop
Morning Sessions
Welcome Kickoff, 09:00-09:15
Conference organizers welcome everyone, lead introductions, and review the day’s plan.
1: AI-A2J in Practice, 09:15-10:30 AM
09:15-09:30: Juan David Gutierrez: AI technologies in the judiciary: Critical appraisal of LLMs in judicial decision making
09:30-09:45: Ransom Wydner, Sateesh Nori, Eliza Hong, Sam Flynn, and Ali Cook: AI in Access to Justice: Coalition-Building as Key to Practical and Sustainable Applications
09:45-10:00: Mariana Raquel Mendoza Benza: Insufficient transparency in the use of AI in the judiciary of Peru and Colombia: A challenge to digital transformation
10:00-10:15: Vanja Skoric, Giovanni Sileno, and Sennay Ghebreab: Leveraging public procurement for LLMs in the public sector: Enhancing access to justice responsibly
10:15-10:30: Soumya Kandukuri: Building the AI Flywheel in the American Judiciary
Break: 10:30-11:00
2: AI for A2J Advice, Issue-Spotting, and Engagement Tasks, 11:00-12:30
11:00: Opening remarks to the session
11:05-11:20: Sam Harden: Rating the Responses to Legal Questions by Generative AI Models
11:20-11:35: Margaret Hagan: Good AI Legal Help, Bad AI Legal Help: Establishing quality standards for responses to people’s legal problem stories
11:35-11:50: Nick Goodson and Rongfei Lui: Intention and Context Elicitation with Large Language Models in the Legal Aid Intake Process
11:50-12:05: Nina Toivonen, Marika Salo-Lahti, Mikko Ranta, and Helena Haapio, Beyond Debt: The Intersection of Justice, Financial Wellbeing and AI
12:05-12:15: Amit Haim: Large Language Models and Legal Advice12:15-12:30: General Discussions, Takeaways, and Next Steps on AI for Advice
Break: 12:30-13:30
Afternoon Sessions
3: AI for Forms, Contracts & Dispute Resolution, 13:30-15:00
13:30: Opening remarks to this session13:35-13:50: Quinten Steenhuis, David Colarusso, and Bryce Wiley: Weaving Pathways for Justice with GPT: LLM-driven automated drafting of interactive legal applications
13:50-14:05: Katie Atkinson, David Bareham, Trevor Bench-Capon, Jon Collenette, and Jack Mumford: Tackling the Backlog: Support for Completing and Validating Forms
14:05-14:20: Anne Ketola, Helena Haapio, and Robert de Rooy: Chattable Contracts: AI Driven Access to Justice
14:20-14:30: Nishat Hyder-Rahman and Marco Giacalone: The role of generative AI in increasing access to justice in family (patrimonial) law
14:30-15:00: General Discussions, Takeaways, and Next Steps on AI for Forms & Dispute Resolution
Break: 15:00-15:30
4: AI-A2J Technical Developments, 15:30-16:30
15:30: welcome to session 15:35-15:50: Marco Billi, Alessandro Parenti, Giuseppe Pisano, and Marco Sanchi: A hybrid approach of accessible legal reasoning through large language models 15:50-16:05: Bartosz Krupa – Polish BERT legal language model 16:05-16:20: Jakub Dråpal – Understanding Criminal Courts 16:20-16:30: General Discussion on Technical Developments in AI & A2J
Closing Discussion: 16:30-17:00
What are the connections between the sessions? What next steps do participants think will be useful? What new research questions and efforts might emerge from today?
