Courts are currently developing new policies and issuing guidelines for good AI practices
The rapid adoption of generative AI in Ontario’s legal system is forcing courts to take steps to address the technology, including issuing updated practice guidance.
During a recent webinar on Thursday, June 18, Justice Peter Lauwers of the Court of Appeal for Ontario, Justice Fred Myers of the Ontario Superior Court of Justice, Justice Jill Presser of the Ontario Superior Court of Justice, and Justice Joseph Kenkel of the Ontario Court of Justice discussed the pros and cons of AI and speculated about how its use will evolve. Overall, while the justices recognized that AIs could be problematic, especially if used improperly or by people with a limited understanding of either the technology or the legal regime, they valued the possibilities AI holds for improving access to justice.
Kenkel, for example, has found that self-represented parties before the courts present their cases more effectively after employing AI tools. “Their understanding of the legal issues that they discuss, the questions that they ask, and their final submissions, they’re all much, much better,” he said.
In a recent case, Kenkel said that he even suggested that opposing sides, including a self-represented party, use AI tools to translate some key documents. That approach was so successful, he said, they were able to skip directly to stage two during a s. 278 application under the Criminal Code.
All the justices, however, were adamant that everyone participating in the justice system be careful to avoid the pitfalls AI can create, including hallucinations and fabricated evidence, such as deepfakes or made-up quotations. This is especially true for lawyers, who serve as officers of the court, said Myers.
“Our system is predicated on the integrity, not just of the system, but of the system’s players. We have no army to enforce our judgments. The population takes our judgments as authoritative because they have faith in the fairness and the neutrality and the independence of the process, and lawyers are part of that process.”
Lawyers must adhere to the highest standards when using AI because the courts have very few tools that would prevent abuse. Myers pointed to the United States as an example where even sanctions have proved ineffective in deterring improper AI use by legal professionals. And he noted that AI misuse is on the rise in Canada, citing data compiled by access-to-justice watchdog Courtready.
But when AI is used ethically, it promises to offer a boon to the province’s courts. In advocating for its benefits, Presser cited a highly successful Quebec Superior Court pilot project that created bespoke tools for its judges and provided them with extensive training in using the technology. She reported that 84 percent of participating judges found AI compatible with judicial function, and 89 percent wanted to keep using it.
She said she hoped a similar pilot project would be run in Ontario in the near future and sees the adoption of technology as vital to the efficiency of the court system.
“I actually think the justice system owes it to the public to explore the benefits that can be gained from these tools, because we owe the public the most efficient and effective human justice system possible.”
The catch, she explained, is to clearly explain how they are using the tools and to emphasize that human judges, relying on human judgment, will always be what the public can expect.
“I do think it's very important for courts, as we start adopting AI, to be very transparent with the public about the use of AI – to communicate to justice system users and the public that we will be using AI. ‘Here’s how we'll be using AI’, but very crucially, ‘here’s how we will not be using AI.’”
One way Myers would like to see AI used is as an automatic hallucination-detection tool for any document submitted through the court’s document-submission portal. He hoped something like that would save countless hours by fact-checking submissions.
Caution was on the mind of Lauwers, who referred to studies that suggested that people who resort to an AI-first approach to solve problems or answer questions, without trying to develop an understand of the subject matter first, will have a poor understanding of the problem and will not retain memory of what the AI-suggested answers are, but that users who use AI to validate their own research and understanding will create improved output, which is the approach he wants judges to take, if they decide to rely on AI supports.
He was also concerned about deepfakes and the influence they can have. Again, referencing academic research, he said that audio and visual deepfakes are highly persuasive and pose the risk of biasing those who see them, including members of the court.
“Remember the expression ‘unring the bell’? Well, it turns out you can't unring the bell; even if you're told it’s fake, you will still let it influence your decisions, so the question is: who is to decide whether something is deepfake or not? The argument would be that it shouldn’t be the judge who’s trying the case, and it shouldn’t be the fact finders,” adding that an exercise in judgment should probably occur before it gets in front of the judge.
Deepfakes were also a key concern for Presser, who said that in family law cases, deepfakes and fabricated evidence, such as AI-generated text chains full of threats, harm, or kidnapping, are becoming a serious problem.
“I think the question of the authenticity of evidence, or whether it's deep-faked AI-generated evidence, is an issue that is going to only grow in our courts and in all areas.”
To deal with all of the issues and challenges posed by AI, the courts are working hard to ensure their members know how to respond to AI and how to incorporate it best.
Lauwers, for example, is deeply involved in that process as chair of the civil rules committee and as notional chair of the tri-court committee, which is a committee comprising members from the Court of Appeal, the Superior Court, and the Ontario Court.
As part of his civil rules committee duties, Lauwers said that lawyers will soon receive a consultation package being sent to the profession. (He said that some people may have already seen an earlier version.) It addresses the rules on evidence for civil cases, building on proposed AI-evidence changes the subcommittee floated last year. Lauwers noted that the committee would like to hear comments and opinions on its AI proposals.
Among the topics addressed will be the “critical distinction between evidence generated using acknowledged generative AI and evidence generated using unacknowledged AI, so we expect folks to acknowledge their use of generative AI in creating evidence.”
“Acknowledged AI-generated evidence,” a term credited to Maura Grossman, a research professor at the University of Waterloo, refers to content that was enhanced or improved by AI systems that have good science behind them and are open to scrutiny. These can include tools used by accident reconstruction investigators or probabilistic genotyping DNA researchers.
One of the many changes that will also be a part of the package will include one that addresses the certification of quotations “because we’ve discovered that sometimes…the quotes themselves are not accurate.” That concern is not hypothetical: Justice Fred Myers recently referred a lawyer to the Law Society of Ontario over a factum that cited real cases but contained fabricated quotations.
It will also create a system whereby parties can challenge evidence as being AI-generated. “It’s a bit of a flip on onus, the same way as you would see in a limitations case,” he explained.
The work the tri-committee is doing examines how judges can and should use AI themselves. To date, he explained, the only policy addressing this issue is the guidance issued by the Canadian Judicial Council. The committee is using that as a foundation, including its need for humans to always be in the loop. Another key plank will be that “no judicial function is delegated to artificial intelligence.”
The Ontario Court of Justice has taken the basics of that policy and issued an interim, eight-page guidance “with respect to proper uses of AI to inform judges, but also everybody who works with the judges.” In particular, he said the court needed to ensure that the nearly 300 justices of the peace were supported as they dealt with the ever-evolving technology.
While the documents are not available to the public, he said they involve three core values: human decision-making, data safety (using only Microsoft Copilot, keeping data in Canada, etc.), avoiding algorithmic bias, and ensuring that there is representative data in any AI application.
“Here’s our bright line rule for judges and justices of the peace: if the task determines or shapes the legal outcome, the assessment of credibility, or contested facts… do not use generative AI.”
Presser also went into some detail about practice directions issued in February 2026, by the Superior Court. She said there are “slight variations” for each area of practice (family, civil, criminal), but they all set out a commitment to maintaining the integrity of the justice system, and the fact that that is a shared responsibility of all justice system participants.”
