The Civil Rules Committee's draft rules cover AI-generated evidence and deepfakes, with rule text expected in November
Ontario is preparing civil procedure rules to govern AI-generated evidence in court, responding to concerns that hallucinated citations, AI-altered quotations and passive reliance on AI tools are already creating problems in Canadian courtrooms. Rather than restricting AI use outright, the proposed rules take a judicial-control approach – demanding transparency and scrutiny while leaving prompt disclosure to the judge's discretion.
The rules were developed by the Ontario Civil Rules Committee (CRC), chaired by Justice Peter Lauwers of the Court of Appeal for Ontario. In 2025, the committee’s subcommittee on artificial intelligence, which Lauwers also chairs, sought feedback from the profession. After consulting, the subcommittee returned to the committee with what Lauwers describes as a “seriously amended set of rules,” which have been submitted to Attorney General Doug Downey. Revised rule text is expected in November, Lauwers says.
"Courts cannot prevent AI's arrival, but they can establish appropriate safeguards to protect the integrity of adjudication and the development of the common law," Lauwers wrote in a summary of the proposed rules provided to Law Times in October.
The first pillar covers acknowledged AI-generated evidence – material disclosed as having been produced, in whole or in part, by an AI system. A court may require the proponent to identify the AI system used, its training data, inputs and prompts, and supporting evidence of validity and reliability. Two forms of bias are priority concerns: sampling bias and discriminatory bias of the kind prohibited by the Ontario Human Rights Code.
The second rule addresses unacknowledged AI-generated evidence, including deepfakes. Because fabricated video or audio can be compelling and difficult to set aside, the proposed rule adopts a presumption of inadmissibility once material is found to be undisclosed AI-generated content.
"The premise is that judicial control, rather than prohibition, is the practical response to AI," Lauwers wrote. "Courts should demand enough transparency to test AI-generated material while preserving privilege and preventing fabricated evidence from distorting the adjudicative process."
The proposed rules flag prompt disclosure as a particular concern. The way a question is framed can materially bias an AI output. Lauwers uses the example of a prompt asking an AI to compare someone’s productivity in the workplace with their peers versus asking it to compare their lack of productivity – the prompts would produce a materially different result.
Prompts may reveal litigation strategy and are arguably protected by solicitor-client privilege, so the rule would not compel prompt disclosure as of right and instead leaves it to the judge’s discretion. "We're not going to make you disclose your prompts, but if you don't, we may not admit the evidence because the evidence may be so distorted by the prompt that it's unreliable or invalid," Lauwers says.
The problems the rules address are already arriving in Ontario courtrooms. Lauwers says the most common appellate-level issues are hallucinated citations and AI-edited quotations. “We’ve changed the rules a few times to require people to certify … the citations that they make and the material they submit,” he says.
He flags passive reliance on AI as the broader risk, citing a 2025 MIT study showing that users who start with AI rather than their own reasoning often cannot explain the answer minutes later – a pattern he has observed in counsel pressed from the bench.
Kathryn Manning, a litigation partner at DMG Advocates in Toronto who sits on the CRC's AI subcommittee with Lauwers and chairs the OBA's technology and practice innovation committee, says the same dynamic is visible at the trial level. Manning says a small claims court judge recently told her that AI-generated materials are arriving with increasing frequency – better written in some cases, but raising accuracy concerns.
The concern extends beyond Ontario: British Columbia Court of Appeal Justice Susan Griffin recently warned that courts "cannot provide an unlimited audience" to AI-generated submissions.
Beyond the evidentiary rules, Ontario's civil procedure reform is also reshaping the discovery landscape. The Civil Rules Review Working Group's final policy report, released in December 2025, proposes moving from a relevance-based disclosure model to a reliance-based one, in which parties would produce documents they rely on and those they know the other side would need.
"I don't think it reduces [the e-discovery burden] necessarily, because you're still going to collect documents from your client, and you need to figure out what you're going to rely on," Manning says. "I think it actually will move the e-discovery burden earlier in the case," she says.
The proposed scaling back of oral examinations – from elimination to approximately three hours in most cases – also concentrates weight on the document record. "If you can ask fewer questions, then you will rely on documents a little bit more than maybe getting admissions," Manning says. "The document will have to be the admission in some instances."
Standards for authentication of AI-generated records in discovery are not yet settled, Manning says, but the forthcoming rules will address this. "Ontario took a very practical, measured approach to the issue," she says, contrasting it with other Canadian jurisdictions that issued quick blanket restrictions on AI use in court.
The proposed rules are not yet law, but Lauwers says they break no new ground. "They don't mark any kind of development in the true sense of the word," he says. "Putting forward evidence that is not reliable or valid is an abuse of process."
Lauwers has advised both Superior Court and Ontario Court judges to use the draft rules as a practical guide. "Judges know about it. The lawyers don't, except for those who've been involved in the consultation process," Lauwers says.
The Canadian Judicial Council also released 2026 guidelines on evaluating digital media evidence, flagging the "liar's dividend" – the risk that authentic evidence will be challenged as fabricated.
Lauwers says he hopes to finish the process "in the next month or so," at which point the standards will become publicly available to the profession.
Regardless of the rules in their specific province, Manning says lawyers should get up to speed on technological developments in the area: "I think we have a duty to be competent in everything we do in legal practice, and that involves understanding the benefits and the risks of certain technologies that are not going anywhere.”
