Artificial Intelligence, Access to Justice, and the Future of Law

Here is a speech I prepared for the JusticeNet Advisory Board Meeting – June 8, 2026 

Introduction

For more than half a century, access to justice has been one of the most persistent challenges facing legal systems in Canada.

Governments have funded legal aid. Courts have simplified procedures. Law societies have promoted pro bono services. Legal clinics have expanded their reach. Yet despite these efforts, legal services remain unaffordable for many people and legal processes remain inaccessible to many Canadians.

In Hryniak v. Mauldin,[1] Karakatsanis J. writing for the Supreme Court of Canada stated

“Ensuring access to justice is the greatest challenge to the rule of law in Canada today.”

Chief Justice Richard Wagner in his well-known speech Access to Justice: A Societal Imperative noted that “Access to justice can mean many things.” Speaking in 2018, in the age when search engines were the general public’s compass to finding legal information, and well before the dawning of the age of AI, he noted a key barrier to access to justice is both a lack of legal information and too much misinformation.

A third barrier to access to justice is lack of access to legal information. How many problems could be avoided if the public had a higher level of legal knowledge, or at the very least quick and affordable access to basic advice?

On the other hand, a little knowledge is a dangerous thing, and that’s no more evident than when you see a self-represented litigant in court, relying on some arcane point of law that she Googled, without realizing why it doesn’t actually help her. Or without noticing that everyone else in the room is getting frustrated at the waste of time.

It isn’t just a problem of a lack of information; there is also too much misinformation…

Most discussions about AI focus on lawyers, law firms, and productivity. I want to focus today on the benefits and challenges of AI for the public that cannot afford lawyers or qualify for legal aid.

A spoiler alert – we are seeing many self-represented individuals, paralegals, and poorly resourced lawyers in Canada increasingly relying on AI to help understand their rights and to advance their cases before courts and tribunals.

However, they lack an understanding about the inherent unreliability of large language models (LLMs), also known as generative AI models. As a result, their legal submissions and evidence are being rejected by courts and tribunals. Courts and law societies are also holding self- represented litigants to the same ethical standards as lawyers, which though understandable, is creating further problems for them.

The Future of Self-Represented Litigants

One of the most visible symptoms of the access-to-justice crisis has been the growth of self-represented litigants. As you know, they do so because they cannot obtain affordable assistance.

AI may significantly improve self-represented litigants’ ability to navigate legal processes. AI can help them do many things including understand procedures; organize documents; prepare chronologies; draft forms; identify issues; and understand and locate judicial decisions.

It doesn’t put them on an equal footing with experienced counsel with more advanced AI capabilities. But it helps the David’s take on the Goliaths (both corporate and institutional) with highly intelligent AI slingshots.

AI generated chatbots like ChatGPT, Gemini, Claude, and Meta AI are far more user-friendly that traditional Google search. Rather than getting a myriad of search results, litigants can get information organized and analyzed, complete with authoritative looking sources like case citations and articles. Litigants can get the LLMs to help determine what claims or defences they have, write demand letters or submissions to tribunals; help develop cross-examination strategies; and to help with evidentiary issues.

AI may also improve access to justice by making court and tribunal internal operations more efficient.[2]

These are important and welcome developments.

Unknown risks in AI

A serious problem is that self-represented litigants do understand how LLMs are trained and that they have inherent reliability limitations.

LLMs generate responses to prompts by identifying statistically likely outputs based on patterns learned from training. The results appear authoritative, even if they are false, misleading or simply made up, what are often referred to as hallucinations”. I can personally attest to this. They do not merely produce a Googled list of sources litigants can review and try to make sense of. The responses produced by LLMs are seductively reliable, which is why we have seen a huge uptake in their use by lawyers including in large firms, by self-represented litigants and their agents and paralegals.

Even trained lawyers are falling prey to relying on LLM generated outputs without verifying their accuracy. This has led to an epidemic of cases around the world where lawyers before tribunals and courts and have cited non-existent cases, cited cases for propositions they do not stand for, or have used quotations from authorities that do not exist or are inaccurate.

Even adjudicators like judges and arbitrators have been caught relying on non-existent cases. The most egregious example is an arbitrator whose decision was recently overturned by a Quebec court because of the use of LLM fabricated reasons.[3]

AI hallucinations and the self-represented and poorly resourced litigant

Self-represented litigants, paralegals, and poorly resourced lawyers are unknowingly finding themselves before courts and tribunals and finding their strategy of using LLMs has prejudiced their cases. Individuals and paralegals may not understand the rules. Some lawyers are extremely busy and rely on outside freelance lawyers or juniors and do not take the time to review what they get back to check for hallucinations. This is a common theme in the reported cases.

Self-represented litigants, paralegals, and poorly resourced  lawyers are coming up against legal and ethical rules designed to maintain the rule of law and evidentiary admissibility rules they didn’t know about, realize they were breaching, or were too busy or careless to comply with. The result, the litigant’s hopes of successfully getting success to justice using LLMs as their proxy legal advisors or agents are stymied because they have not understood how AI generated outputs can be unreliable and what courts and tribunals and law societies expect of professional and non-professional litigants.

To be clear, I understand the requirement to preserve the integrity of the legal system and the rule of law. But I wonder if more needs to be thought about how these rules affect individual litigants and how the ethical rules and access to justice imperative can be balanced.

In short, we not only have a slow adoption of AI problem in Canada, we also have a knowledge and skills problem that is affecting justice by the most vulnerable.

There are a litany of examples, but let me give you an illustration of decisions rendered in the last few months only.

• Decision353/24R, 2026 ONWSIAT 374. Paralegal acting for worker in appeal before Ontario Workplace Safety and Insurance Appeals Tribunal. Citing non-existent cases. Tribunal refusing to rely on cases. Appeal dismissed.[4]
• Whitaker The Owners, Strata Plan VIS1144. Self-represented litigant relying on on-existent cases. Court not relying on arguments based on those cases.[5]
• Blanes v. Stellers Jay Academy Inc., 2026 BCCRT 836. Self-represented litigant relying on on-existent cases. Court not relying on arguments based on those cases and noting the court rule that litigants cannot reply on AI generated evidence or materials.[6]

CRT rule 8.3(8) and (9) prohibit parties from including nonexistent cases or legislation, or false or fabricated evidence, including material created by artificial intelligence.

• Lee AutoCanada MR Motors GP Inc., 2026 BCCRT 854. Self-represented individual in contract claim relying on materials produced by LLMs excluded for being unreliable.

Mr. Lee relied on information generated by ChatGPT and Claude, generative artificial intelligence(AI) tools. Previous CRT and court decisions have declined to rely on AI-generated evidence, because of its inherent unreliability. See, for example, Bahia v. ICBC, 2025 BCCRT 453. He also relied on statements from internet forums or message boards. Although the CRT has the authority to accept hearsay evidence, I find this evidence is not reliable because neither the authors nor their qualifications are identified. So, I give this evidence no weight.

• Iida v. Meadow Lane Equine Clinic Ltd, 2026 BCCRT 626. Negligence claim in which AI generated evidence held unreliable and inadmissible.

Ms. Iida also relies on information generated by ChatGPT, a generative artificial intelligence (AI) tool. She says she asked ChatGPT to summarize the common signs of a horse needing their teeth floated. She also asked ChatGPT to summarize Dr. Wise’s findings, and to review Meadow Lane’s records.

Meadow Lane says that this evidence is not reliable. I agree. Previous CRT and court decisions have declined to rely on AI-generated evidence, due to its inherent unreliability. See, for example, Bahia v. ICBC, 2025 BCCRT 453, and Floryan v. Luke et al., 2023 ONSC 5108. I agree with this approach, and I place no weight on Ms. Iida’s ChatGPT evidence.

A2500445 (Re), 2026 CanLII 47828 (BC WCAT). Appeal to British Columbia Workers’ Compensation Appeal Tribunal. Appeal by worker to the British Columbia Workers’ Compensation Appeal Tribunal. Appeal allowed, but court noting problems with uses of AI producing errors for which worker would be responsible.

I agree with A2501051 (Re), 2025 CanLII 97422 (BC WCAT) that Tribunal decision makers have an obligation to provide sufficient reasons for their decisions, but decision makers at some tribunals … have concluded that this duty does not include the obligation to respond to submissions concocted by artificial intelligence which have no basis in law. Therefore parties who rely on artificial intelligence should be aware that their arguments may not be addressed if they are not based in law.

Although it did not affect the outcome of this appeal, the worker should be aware that, if he is using artificial intelligence, that the use of artificial intelligence in legal processes has numerous potential pitfalls that could be disadvantageous to him…

Further, from my experience, submissions provided using artificial intelligence are relatively easily identified by an experienced decision maker as they frequently contain errors in their native form. Such errors can be disadvantageous to a party if they are not edited for rectification because ultimately the party is responsible for the content generated through artificial intelligence and submitted to WCAT, even if it is incorrect. While WCAT currently does not have any prohibition concerning the use of artificial intelligence, the worker should be aware that if he is using such technology its usage is often easily identified by decision makers, can contain inaccurate information, and can generate poor submissions that are unnecessarily lengthy, contain tone issues, and do not articulate the arguments in a logical manner.

Okogun v. Canada (Citizenship and Immigration), 2026 FC 722. Self-represented litigant in immigration appeal citing non-existent case law in a judicial review that was dismissed. Court holding litigant to same standard as lawyers.[7]

This responsibility [Court AI Notice] applies regardless of whether a party is represented by counsel or self-represented because the obligations on the Court to maintain the integrity of judicial proceedings, safeguard public confidence in the justice system, and uphold the rule of law, are constant, and do not depend on whether a party is represented.

• KH (Re), 2026 ABCJ 83. A family law case where a lawyer used AI in the preparation of her written closing submissions and it hallucinated cases that did not exist. The court reiterated that the court rules related to the use of AI applied both the lawyers and self-represented litigants.[8] Quoting from a prior decision of the Alberta Court of Appeal it stated:

However, counsel and self-represented litigants should not expect leniency where they have failed to adhere to clear and unambiguous requirements.

Confidentiality and privilege concerns

Lawyers and clients need to understand that confidentiality and privilege rights can be lost with improper uses of AI. Before specific information is entered into an AI system, litigants and their lawyers and paralegals must understand how the tool processes, stores, uses, retains, and discloses inputs and outputs.

We are already seeing caselaw in the address when uses of AI might affect these rights. The law is still developing, but litigants and their representatives may be exposing themselves to risks without realizing it.

“ Uploading confidential documents into an open-source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege, and any such conduct might itself warrant referral to the regulatory body and should, in any event, be referred to the Information Commissioner’s Office.” UK and R (on the application of Munir) v Secretary of State for the Home Department (AI hallucinations; supervision; Hamid),  [2026] UKUT 81 (IAC)

“Generative artificial intelligence presents a new frontier in the ongoing dialogue between technology and the law. Time will tell whether, as in the case of other technological advances, generative artificial intelligence will fulfill its promise to revolutionize the way we process information. But AI’s novelty does not mean that its use is not subject to longstanding legal principles, such as those governing the attorney-client privilege and the work product doctrine. Because Heppner’s use of Claude fails to satisfy either of these rules, the AI Documents do not merit the protections Heppner has claimed.” United States v Heppner, 2026 WL 436479 (SDNY 17 February 2026)

“AI is forcing litigants and courts to confront difficult questions about how and to what extent longstanding protections will apply when parties use AI to assist them in the litigation process.” Morgan v V2X, Inc, 2026 WL 864223 (D Colo 3 March 2026) (pro se litigants can use AI without waiving work product protection. “Here, like in Warner v Gilbarco, Inc, 2026 WL 373043 (ED Mich 10 February 2026), Plaintiff can assert work product protections in connection with his AI use. It is true that AI systems like ChatGPT, Claude, Gemini, and others widely available to the public, collect user data for training and other purposes. But in this Court’s estimation, that does not eliminate all expectations of privacy or automatically waive protections.”

Are self-represented litigants being outgunned by enterprise uses of AI tools?

Many law firms and government departments are using sophisticated legal tools in their litigation practises including well-known tools such as Co-Counsel, WestlawAI, Walter, as well as LLM tools that integrate into other applications such as iManage, Co-Pilot and Claude for Legal.[9]

These developments, which will continue unabated, will likely also continue to tip the tables against self-represented litigants or poorly resourced legal advisors.

Use of AI in administrative decision-making

AI is also being used by the courts as an aid to decision-making. One tool that has been subject to significant litigation is the use of an AI tool called Chinook in immigration cases. A summary of the caselaw is in a decision called. Espinosa Cotacachi v. Canada (Citizenship and Immigration).[10]

I am not an expert in this area, but it seems to me that individuals and their representatives do not have the confidence that officers are really deciding the cases themselves based on their own analysis. They raise procedural fairness arguments. But, as applicants have the burden of proof they have not been able to discharge their burdens in judicial reviews.

This raises a question as to whether officers should have the burden of explaining how their decisions were made, that they in fact made them assessing all of the evidence, and not only that there was a human in the loop when making decisions.

It is generally accepted that governments and organizations that use automated decision systems including those AI systems that also assist in decision making that make consequential decisions affecting individuals should disclose to the public how they are used and be able to provide explanations as to how those decisions were made. But, given all the litigation related to Chinook I wonder whether these fundamental principles are being adhered to.

Can JusticeNet help?

JusticeNet may focus today on being a reduced-fee network. But, could it evolve to facilitating an AI-enabled access-to-justice platform? For example by:

• Providing educational resources to it’s network of lawyers on how AI can make their practices more efficient and reminding them about the need for responsible uses of AI.
• Helping legal aid clinics to adopt AI tools  that are secure and confidential to streamline operations to become more productive and cost effective.
• Helping lawyers provide practical advice and guidance to clients in responsible uses of AI.
• Work with law societies, bar associations, courts, tribunals, and governments to provide more information about responsible AI use by members of the public, paralegals, sole practitioners and others.

Concluding Remarks

AI may prove to be one of the most important access-to-justice developments of our generation. If deployed responsibly, it may reduce costs, expand access to legal information, improve legal service delivery, and help more Canadians understand and enforce their legal rights.

But, individuals still need significant help. They understood Google, to a large extent. But, they often do not understand the frailties of LLMs and the requirements of the legal system that imposes ethical standards on them they are not aware of that apply to everyone engaging in the legal system.

Poorly resourced lawyer also need help.

Perhaps there is a new role for JusticeNet to help overcome these challenges.

[1]           2014 SCC 7

[2]           See, OECD, Governing with Artificial Intelligence.

[3]           See, Barry Sookman, AI Ethics: Court Overturns Arbitration Decision

[4]           [25]      The worker’s representative also referenced «Under the HRTO» and section «5.5». Ireviewed the HRC and there is no section 5.5. I cannot consider a section that does notexist.

2.1 Standards of Conduct

2.2 Representatives who appear in a hearing must be licensed by the Law Societyof Ontario. Or they must be authorized to provide legal services in accordancewith the Law Society Act and its regulations and by-laws, in particular theexemptions listed in By-Law 4. All representatives are expected to represent their clients honestly – they must not share information known to be untrue, or assist or encourage a party to misrepresent the facts

• be knowledgeable about the legislation that is relevant to the case – this includes the Workplace Safety and Insurance Act, 1997
• prior legislation that applies to accidents before January 1, 1998
• related regulations and statutes

[5]           In her reply submissions, Ms. Whitaker refers to numerous cases that either do not exist or do not stand for the legal bases that she presents. I find these are likely «hallucinations», meaning false or misleading results generated by artificial intelligence. In AQ v BW, 2025 BCCRT 907 at paragraph 16, a CRT vice chair found that the CRT’s obligation to provide sufficient reasons did not require it to address arguments with no basis in law. I agree with this reasoning. While I have reviewed all of Ms. Whitaker’s submissions and evidence, I have only addressed what is relevant in my decision below.

[6]           In her submissions, Ms. Blanes relies on cases that either do not exist (S.J. v. N.N., 2012 ABQB 75, Zaky v. 2285771 Ontario Inc. o/a Playtime Bowl & Entertainment, 2021 ONSC 5532) or that do not support the legal principles she presents (Housen v. Nikolaisen, 2002 SCC 33). I find these authorities are likely «hallucinations,» meaning false or misleading results generatedby artificial intelligence.

[7]           [22] I agree with the Respondent that Mr. Okogun’s motion to strike the Kaur Affidavit must be dismissed, substantially for the reasons provided by the Respondent in the Response Brief in their responding motion record. Starting with the issue of AI-generated content, Mr. Okogun answered affirmatively when asked by the Court at the oral hearing if he used AI to prepare his materials.

[8]           [276] The Courts in this province have made efforts to ensure that the legal profession and the public understand that caution and care must be taken when relying on Artificial Intelligence in court submissions. These efforts include the issuance of a formal Notice to the Public and Legal Profession on October 6, 2023, entitled Ensuring the Integrity of Court Submissions When Using Large Language Models (the «Notice»). The Notice, which applies to lawyers and self-represented litigants, is signed by the Chief Justices of all three Courts in Alberta and is readily available on the websites of all three Courts. The Notice reinforces the Courts’ commitment to the integrity and credibility of legal proceedings by, among other things, making it clear that there must be meaningful human oversight when Artificial Intelligence is used to prepare court submissions:

Human in the loop: In the interest of maintaining the highest standards of accuracy and authenticity, any AI-generated submissions must be verified with meaningful human control. Verification can be achieved through cross-referencing with reliable legal databases, ensuring that the citations and their content hold up to scrutiny. This accords with the longstanding practice of legal professionals.

The consequence of failing to adhere to the October 2023 Notice is within the discretion of the panel or the individual judge involved with the matter. However, counsel and self-represented litigants should not expect leniency where they have failed to adhere to clear and unambiguous requirements. In most situations, courts will likely consider remedies available under the Rules, including striking submissions or imposing some form of cost award against the party or counsel who failed to follow the requirements of the October 2023 Notice. A court may also determine that a penalty should be imposed, contempt proceedings should be initiated, or that a referral to the Law Society of Alberta is warranted. Maintaining the integrity and credibility of court processes justifies the imposition of proportionate and meaningful sanctions.

[9]           Online: https://claude.com/blog/claude-for-the-legal-industry

[10]          2024 FC 2081