November 20, 2026
AI systems that source content from the open internet can produce responses that are legally inaccurate, and jurisdictionally wrong. If you don’t ask the right questions, or if the AI system isn’t programmed to answer in plain language, the answers can fail to help the people who most need guidance. These are serious concerns, and People’s Law School shares them.
People’s Law School has built and operates a purpose-built AI system designed from the ground up with legal accuracy, human oversight, jurisdictional specificity, and plain-language accessibility as foundational requirements.
What follows is People’s Law School’s position on the responsible use of generative AI in public legal education. We’ll explore the access to justice context, the tools we’ve built, and principles that we consider essential for the development and deployment of responsible AI.
The question we address is not whether AI is safe enough to substitute for legal guidance. It is whether AI, deployed responsibly, can help the hundreds of thousands of British Columbians who have no access to legal guidance at all.
1. The Access to Justice Context
Two thirds of Canadians who face a legal problem attempt to resolve it without a lawyer. When they seek help, they turn most often to the internet. What they find there is, at best, uneven: a mixture of general information, out-of-jurisdiction resources, commercial services with a financial interest in the advice they offer, and, increasingly, AI-generated summaries of uncertain provenance and accuracy.
The question that public legal education must answer, then, is not whether AI is safe enough as a substitute for legal guidance. It is whether AI, deployed responsibly, can help the people who have no access to legal guidance at all.
People’s Law School was founded in 1972 on the premise that the public deserves access to accurate, plain-language legal information — information that helps people understand their situation, know their rights, and take informed next steps. For over fifty years, we have provided that information at no cost to British Columbians, through websites, publications, webinars, a telephone service, a Q&A service, and in-person community outreach.
The emergence of generative AI has not changed that mission. It has changed the landscape in which we pursue it, and, if approached carefully, it has expanded the tools available to us in doing so.
2. Legitimate Concerns About Generative AI in Legal Contexts
The most common concerns fall into the categories set out below.
Accuracy and hallucination
Large language models can produce responses that are confident, fluent, and wrong. In legal contexts, this is not merely an inconvenience.It can cause material harm. A person who acts on incorrect information about their tenancy rights, their entitlement to benefits, or their immigration status may suffer serious consequences.
Jurisdictional errors
Law is jurisdiction-specific. An AI system trained on legal content from multiple countries, provinces, and states will routinely produce responses that mix jurisdictions — citing legislation, procedures, or rights that do not apply in BC.
The erosion of professional judgment
Practising lawyers have reported an increase in clients arriving at intake meetings with AI-generated legal assessments that may need correcting before substantive advice can begin. With AI-generated responses so readily available, clients may start to substitute machine judgment for a lawyer’s wisdom. The concern that AI may undermine, rather than support, the professional relationship between lawyer and client is legitimate.
The replacement of human connection
Law, like medicine, involves more than the transfer of information. A good lawyer reads a client’s situation in full: the spoken and unspoken, the stated and the implied. The concern that efficiency-driven technology may erode that human dimension of professional practice deserves to be taken seriously.
Some voices within the legal profession are already engaging with these concerns constructively. Writing in BarTalk, the official publication of the Canadian Bar Association, BC Branch, lawyer Lisa Picotte-Li argues that AI must be governed by principles that keep human judgment paramount, and calls for governance frameworks, data stewardship policies, accountability mechanisms, and ongoing professional development focused on ethical technology adoption (“New Tools, Old Duties: Artificial Intelligence and the Practice of Law,” BarTalk, June 2025). These are principles that People’s Law School has embedded in Beagle+, our AI-powered chatbot, from the outset.
3. The Critical Distinction: Generic AI vs. Human-Centred AI
The concerns described previously apply, in varying degrees, to general-purpose LLMs, like Claude or ChatGPT. They do not apply, or apply very differently, to purpose-built AI systems designed with legal accuracy, human oversight, and jurisdictional specificity as foundational requirements.
This distinction is rarely made in current discussions about AI in the legal community. It is, however, the most important distinction to draw. The question is not whether AI should be used in legal contexts. The question is: what kind of AI, built how, governed by whom, for what purpose?
The key distinction
Generic AI: trained on the open internet, answers any question from any jurisdiction, no human oversight of legal accuracy, prone to jurisdictional error.
Human-centred AI: grounded in curated, lawyer-reviewed content, from which its responses are drawn directly rather than generated from the model’s own training data; limited to a defined jurisdiction, designed with plain-language accessibility and human oversight as core requirements.
Beagle+ is built on the same class of technology that powers the generic tools described above. What distinguishes it is not the underlying model but rather how humans have intentionally designed its parameters. Beagle+’s answers are retrieved from our own lawyer-reviewed BC content and the system is constrained to answer from that content or to say that it cannot help. A purpose-built system’s curation, retrieval, jurisdictional scope, and human review do the work that the underlying, off-the-rack model cannot do on its own.
4. People’s Law School’s Approach: What We Have Built and How
Every design decision we have made about Beagle+ has been driven by the question: what does a person in a difficult legal situation actually need, and how do we ensure that what we provide is accurate, accessible, and genuinely useful?
People’s Law School’s approach was recognized in 2026 with the Michael Mills InnovAction Award from the US-based College of Law Practice Management. In presenting the award at its Futures Conference in New York on 29 June 2026, the College cited Beagle+’s combination of responsible AI, lived-experience design, and jurisdiction-specific legal accuracy. People’s Law School is the first Canadian public legal education organization — and the first Canadian access-to-justice non-profit — to receive the award.
Human-curated content as the foundation
Beagle+’s content universe is the same content that underpins our website, our publications, and our other services. It is the work of lawyers who are legal educators and writers with deep expertise in BC law and in communicating legal information to non-specialist audiences.
Lawyers review our content and Beagle+ answers
Our website content is reviewed for legal accuracy by volunteer practising lawyers. This is not a one-time exercise: content is updated as legislation, regulations, and case law change.
Equally, Beagle+’s answers are also reviewed by our internal team of lawyers. In the early days after launch, we reviewed all of the conversations for accuracy and helpfulness. As Beagle+’s audience has scaled, we now use another frontier AI as the first layer of review, with humans then doing a second review of selected conversations.
BC-specific and jurisdictionally bounded
Beagle+ is designed and configured specifically for British Columbia. It does not synthesize legal content from other provinces or countries. When a user asks about their rights as a tenant, an employee, or a family member in British Columbia, the response they receive reflects BC law..
Plain-language design
Our content is written in plain language for audiences with no legal background. Beagle+ is designed to explain legal concepts in language that is accessible, practical, and oriented toward helping the user understand their situation and take the next step.
Transparency and honesty about limitations
Beagle+ is explicit about what it is and what it is not. It does not hold itself out as a lawyer, does not provide legal advice, and actively encourages users to seek professional legal assistance where their situation warrants it.
We are also honest about accuracy. Beagle+ is not accurate 100% of the time. But neither are human lawyers. We set a baseline for accuracy prior to public launch (90%), and now calculate that Beagle+ is accurate more than 99% of the time.
Performance monitoring: helpfulness and accuracy
Our own monitoring shows an 83% helpfulness rating across two years of operation, a number that is increasing. This is a measure of user impact, reported by users via a thumbs up and thumbs down button that appears during conversations.
Beyond our own internal metrics, an ongoing research collaboration with the Legal Design Lab and Lift Lab, both at Stanford University Law School, and with assistance from Access Pro Bono, will generate independent empirical evidence of Beagle+’s performance. We look forward to learning how Beagle+ performs against benchmarks for quality legal Q&A.
For accuracy, we score Beagle+’s responses against structured safety and value rubrics developed in-house and applied across substantial test datasets of BC legal questions. Beagle+ responses are assessed for factual accuracy, appropriate risk warning, correct identification of the legal issues in play, jurisdictional correctness, clarity of next steps, tone, and whether the response leaves the user better able to act. Using these rubrics, our team of lawyers reviews Beagle+ conversations weekly, identifying weaknesses and refining the system continuously.
When Beagle+ gets something wrong
No system of this kind is free of error, and no one should trust one that claims to be. What matters is what happens when an error occurs.
Our team has flagged 127 out of over 15,000 Beagle+ conversations that contained an inaccuracy material enough to affect a user’s understanding of their legal rights, or of the steps available to them. In each case we trace the error back to its origin in our source content, make the necessary addition or correction there, and then re-test.
This loop is what makes the system trustworthy over time.. Because Beagle+ answers from our content rather than from the open internet, an error is usually traceable to something our content did not say clearly enough, or did not say at all. Fixing it improves not only the chatbot’s next answer but the underlying resource that thousands of people read directly. The correction propagates.
Performance Since Launch
18,000+ conversations since February 2024 (June 2026 figures)
83% user-reported helpfulness rating across two years of operation
127 conversations (less than 1%) identified in internal review as containing an inaccuracy material to a user’s understanding.
Responses drawn from BC-specific, lawyer-reviewed content
Research collaboration with the Stanford Legal Design Lab/Lift Lab underway — results not yet available
5. Extending the Human-Centred Approach: Multilingual Access
This year, in partnership with Éducaloi (a Quebec-based non-profit organization that provides plain-language legal information) and with support from the Notary Foundation of BC, People’s Law School completed the AI-assisted
translation of our entire website into French — the first public legal education organization in Canada to do so at this scale.
The methodology reflects the same human-centred principles that govern everything we do. Before translation began, People’s Law School and Éducaloi collaborated to develop a glossary of over 400 common English legal terms, concepts, and organization names translated into French, ensuring that the most important and frequently used language was rendered consistently and accurately throughout the site.
We then applied a two-track quality assurance process.
In the first track, English content was translated into French using ChatGPT, and the resulting French text was then translated back into English using Claude. Comparing the original English with the Claude-produced English allowed us to identify systematically where meaning, legal nuance, or clarity had been lost in translation.
In the second track, professional French translators at Éducaloi reviewed representative samples of AI-generated French translations.
Together, the two tracks provide a richer basis on which to assess the accuracy of translation at scale.
A translation quality study now underway with the University of British Columbia is designed to test how much the two-track process actually catches. The AI enabled the scale; the human review, and our willingness to keep measuring it supports the quality.
Francophone British Columbians can now access plain-language legal information across all the topics we cover in English. This is, in essence, the human-centred approach applied to language access: using technology to extend reach, while maintaining the human oversight that makes the information trustworthy.
6. The Access to Justice Imperative
The concerns about AI in legal contexts are most acutely relevant in the context of practising lawyers advising paying clients. In that context, the stakes of error are high, the professional and regulatory frameworks are well developed, and the lawyer’s role as expert advocate is irreplaceable.
Public legal education operates in a different context. Our users are typically not clients consulting a lawyer. They are people who have encountered an everyday legal problem — a dispute with a landlord, an employment issue, a question about a will, an immigration concern — and who have no access to a lawyer. For them, the choice is typically not between a lawyer and an AI chatbot. It is between no information and some information. The quality and reliability of that information matters enormously.
The risk of providing no information to this population is significant.. Tenants who do not know the law governing evictions may not recognize an unlawful notice, let alone challenge it. Workers who do not know their entitlements are unlikely to pursue them.
Used responsibly, AI can provide accurate, plain-language, jurisdiction-specific legal information to people who would otherwise have none, at any hour, in any location, at no cost. That is not a threat to the legal profession. It is a complement to it — and a contribution to the kind of society in which the rule of law has genuine meaning for everyone, not only for those who can afford to access it.
Who uses Beagle+?
British Columbians facing everyday legal problems: disputes with landlords, employment issues, family law questions, estate concerns, immigration matters, questions about welfare and disability benefits. People who, in the absence of a resource like Beagle+, would have no reliable source of legal guidance.
7. Our Principles for Responsible AI in Public Legal Education
People’s Law School offers the following principles, drawn from our own experience, as a framework for the responsible use of AI in public legal education. We commend them to funders, policymakers, partner organizations, and others engaged in this conversation.
Human oversight is non-negotiable. Not only in reviewing AI’s performance, but in laying the groundwork for how AI does its job.
Jurisdictional accuracy is essential. Legal AI tools for public education must be limited to the jurisdiction they serve. Generic, multi-jurisdictional tools are not appropriate for providing legal information to a specific population.
Plain language is a legal access issue. Plain-language design is required for effective public legal education.
Transparency about limitations is mandatory. AI tools must be honest about what they are, what they can do, and where their limits lie. They must not hold themselves out as providing legal advice, and must actively encourage professional consultation where appropriate.
Performance must be measured independently. Claims about AI accuracy and helpfulness must be supported by evidence, not assertion. Independent research and ongoing monitoring are requirements of responsible deployment.
We did not adopt a human-centred approach because it reflected a growing consensus. We adopted it because, in public legal education, trust, accountability, and human judgment are fundamental.
8. Conclusion
The conversation about AI in the access to justice sector is important, and it is one that People’s Law School is committed to contributing to.
The hundreds of thousands of British Columbians who face legal problems without access to a lawyer deserve reliable information. They deserve tools that have been carefully built, honestly described, rigorously tested, and continuously improved. That is what People’s Law School has built. That is what we remain committed to.
About People’s Law School
People’s Law School is a non-profit public legal education organization founded in 1972 and based in Vancouver, BC. We provide free, plain-language legal information to help British Columbians understand their rights and resolve the legal challenges of everyday life, through our websites (peopleslawschool.ca, dialalaw.ca), the Beagle+ AI chatbot, Dial-A-Law, webinars, a Q&A service, print publications, and in-person community outreach. We are funded primarily by the Law Foundation of BC and the Department of Justice Canada, with project funding by The Notary Foundation of BC.
Contact
Patricia Byrne, BA JD MBA
Executive Director
People’s Law School