Lawyers say the rulings remind them of their existing duties, rather than introduce new obligations
In early 2024, a British Columbia court flagged Canada’s first-known case of a lawyer submitting filings with AI-hallucinated case law. In the two-plus years since that first decision in Zhang v. Chen, the issue it highlighted has only become more widespread: courts and tribunals across Canada have issued at least 150 more decisions addressing instances of lawyers and litigants producing hallucinated case citations or arguments with generative AI tools, according to a widely-cited tracker launched by Paris-based researcher Damien Charlotin.
Those decisions offer some valuable insights. They provide examples of common scenarios and oversights that can lead to lawyers failing to verify their work, helping firms and solo practitioners identify guardrails to prevent the same mistakes. They also show how courts and tribunals are responding to these errors, laying out potential penalties and consequences.
But for some lawyers, these decisions have had only a limited impact on how they use AI tools in their practices. The reason is simple: the courts are not really telling them anything they don’t already know.
“Does it change how I practise personally on a personal level? No, it does not change how I practise,” says Laura Easton, an intellectual property lawyer at Smart & Biggar, of the growing jurisprudence on AI-hallucinated court submissions. “I read the cases that I relied on before the court, before generative AI existed, and I read them now that there’s the ability to use generative AI as a tool.”
The courts have recognized that genAI is “just a tool and the human is ultimately still responsible for their professional responsibility,” says Maya Medeiros, a partner at Norton Rose Fulbright who specializes in intellectual property matters. “You can’t delegate your professional judgment… lawyers already knew this.”
“Just like I couldn’t blame a summer student if they got a case wrong – you’re ultimately responsible as the lawyer or the supervisor,” Medeiros adds. “That can’t be delegated to a tool. You can’t blame a tool for that mistake.”
Case decisions are not the only guidance on responsible AI use that courts and legal regulators have provided to lawyers and litigants. Several courts across Canada have published practice directions urging caution, emphasizing the importance of human review of work produced by AI tools, and encouraging lawyers to be transparent with clients about their use of AI; some of these predate Zhang v. Chen. Law societies have published similar guidelines over the last three years.
However, few of these guidelines say anything materially new, serving more as reminders of principles lawyers are already familiar with, argues Adam Goldenberg, a partner in McCarthy Tétrault’s litigation and dispute resolution group. While such reminders are laudable and helpful for self-represented litigants, they shouldn’t be necessary for lawyers, he says.
“It should be intuitive to anyone practising before any court in Canada that submitting a document that hasn’t been carefully reviewed, and that refers to authorities that do not in fact exist, is a serious lapse in judgment at the very least, if not a breach of professional obligations to the court and to other participants in the justice system,” Goldenberg argues.
Easton notes that the body of court decisions addressing AI-hallucinated case citations or quotes generally flags issues such as abuse of process or miscarriage of justice – concepts that predate AI tools. These rulings have not introduced a “shift in lawyers’ overall duties or obligations to the court,” Easton says. Instead, they have prompted lawyers and law firms to focus on AI literacy.
“The idea is, if your duty is to accurately represent the law before the court, and now you’re relying on something else to try to help you do that, what are you putting in place to ensure that you’re still meeting your duties now that you’re trying to do it with a new tool?” Easton asks.
From that perspective, the courts’ decisions have helped remind lawyers and firms of the importance of guardrails for AI tools. These include establishing workflows to verify the accuracy of AI-produced work and making sure you understand and trust the AI tools you’re using, says Medeiros. The lawyer adds that the rulings have also highlighted the importance of having a remediation plan ready in case something does go wrong.
Medeiros points to Zhang v. Chen, in which the BC Supreme Court declined to order special costs against the lawyer who submitted AI-generated case citations, though the court held the lawyer personally liable for other costs. In declining to order special costs, the court noted that the lawyer had taken steps to correct her errors, apologized to opposing counsel and the court, and did not intend to deceive or mislead the court.
The body of case law on AI-hallucinated citations and quotes has shown that courts are weighing “what corrective actions are taken and what remedial steps are implemented,” Medeiros says. “If you’re proactive about it as soon as you discover the error and try to remedy [it], that might reduce the negative impact overall on everybody involved.”
But the rulings have also raised some questions, Medeiros says: What is the standard of verification that lawyers must meet for work produced with AI? In the event of a mistake, what corrective actions are required? What is the standard for determining whether a lawyer made an innocent mistake versus acting with an intent to deceive? Medeiros also anticipates that questions about how the courts themselves use AI tools will also arise.
In Easton’s personal experience, though, the prospect of hallucinations comes up in conversations with colleagues less than other issues involving AI. Concerns about how AI tools treat confidential information or could potentially result in copyright infringement are more widely discussed, she says.
Goldenberg notes that although there have been many high-profile examples of cases involving hallucinated citations or quotes, in reality, they “represent an infinitesimally small minority of cases that come before the court.” He adds that most of these cases involve hallucinations filed by self-represented litigants rather than lawyers.
While he expects to continue to see both self-represented litigants and lawyers make mistakes with AI before the courts as AI tools become more ubiquitous, he recommends that other lawyers try to withhold excessive judgment as the industry continues to adjust.
“Every time there is a court decision like this, lawyers across the bar at big firms and small firms have two thoughts,” Goldenberg says. “The first thought is, holy smokes, this is a good reminder that we must be careful to ensure that we never file a document in court that contains a hallucination.”
The second thought, Goldenberg says, concerns how many lawyers – regardless of the specific environments they practise in – “suffer from the very human limitations that have always defined law practice for everyone, at every kind of firm and in every type of practice: fatigue, overwork, stress, excessive client demands, resource constraints, limited budgets. It’s those ingredients that, when combined in an environment without sufficient controls, can lead to the sorts of errors that result in lawyers filing hallucinated materials with courts.”
Whenever he reads another story about a lawyer filing hallucinated materials, Goldenberg adds, “My instinct is not initially condemnation, but rather some grace for the lawyers who have erred.” 
