Every family WhatsApp group has that one uncle. You know the one. He forwards a video of a goat speaking fluent Yoruba, adds three crying-laughing faces, and swears on his late grandmother that it is real.

In the family chat, the stakes are low. Somebody’s daughter replies “Uncle, it’s AI,” and life goes on.

In a courtroom, the stakes are very different.

Ontario’s Civil Rules Committee is finalizing new rules of civil procedure for evidence created or altered by artificial intelligence. As Law Times reported on November 3, a revised draft has gone to the Attorney General, and the rules would let judges demand to know which AI system produced a piece of evidence and why anyone should trust it. Undisclosed AI content, including deepfakes, would be presumed inadmissible once it is discovered. The rules are not law yet, but once they are, Ontario litigants will need a very good answer to the question “who, or what, made this?”

What exactly is Ontario proposing?

The work started in 2025, when an AI subcommittee of the Civil Rules Committee, chaired by Justice Peter Lauwers of the Court of Appeal for Ontario, put four proposals out for comment. According to the Law Commission of Ontario’s submission on them, those proposals were:

• A definition of AI written into the Rules.
• Disclosure duties for anyone relying on evidence made wholly or partly by AI: name the software, describe the kinds of data it was trained on, and show the output is valid and reliable.
• A process for challenging fabricated or altered evidence, so suspected fakes no longer enjoy the usual presumption of authenticity.
• A test for AI output in expert evidence, so a machine’s “opinion” faces the same scrutiny as a human expert’s.

After consultation, the subcommittee sent what Justice Lauwers called a “seriously amended set of rules” to the Attorney General. In the latest version, as reported by Law Times, a court could require the party relying on disclosed AI evidence to identify the system, its training data, and the inputs and prompts used, plus evidence of reliability. Bias is flagged as a priority concern, including discrimination under the Ontario Human Rights Code. Prompts are not disclosed automatically, because they can reveal litigation strategy and may be privileged. That is left to the judge.

Can AI-generated evidence be used in court in Canada?

Yes, and that is the point many people miss. Nobody is banning AI from the courtroom. A photo enhanced by software, a transcript produced by speech recognition, or an AI analysis of a mountain of financial records can all be legitimate evidence.

What the draft rules change is the burden. If you want a court to rely on what a machine produced, you have to show your work. Think of it as the evidentiary version of your teacher asking you to show your workings in a maths exam. The right answer is not enough. The court wants to see how you got there.

The sharpest edge is reserved for evidence that was AI-made but never disclosed. Once that is uncovered, the presumption flips: the material is presumed inadmissible. In plain English, if you try to sneak the talking goat past the judge and get caught, the goat goes home.

Don’t we already have rules for this?

Some lawyers think so. In 2025, the Toronto Lawyers’ Association questioned whether the changes were needed at all, pointing out that Ontario’s Evidence Act and the Canada Evidence Act already contain safeguards for electronic records.

It is a fair question, and here is my answer. Those safeguards were built for a world where the worry was whether an email had been tampered with or a scanned contract was complete. They assume a record starts out real and then might be changed. Generative AI breaks that assumption. A deepfake is not a tampered record. It was never real in the first place.

Ontario courts have also already had to deal with AI going wrong in a different way. Since December 2024, under amendments to the Rules of Civil Procedure, lawyers must certify that the authorities cited in their factums are authentic. Fake, AI-invented case citations have already caused real trouble: in Ko v. Li, 2025 ONSC 2766, a lawyer who cited non-existent cases was called to a contempt hearing. Fake cases were the first act. Fake evidence is the sequel, and sequels are rarely better.

Where the draft rules could trip up

I support where Ontario is heading. But after years of turning shiny technology into things people actually use, I know good ideas can die in implementation. Three things worry me.

Cost. Proving that an AI tool is “valid and reliable” sounds simple. In practice, it can mean hiring an expert. The Ontario Bar Association warned that this could be too heavy for ordinary witnesses, and suggested pre-approved tools or reusable vendor certificates. I like that idea. Without something like it, a well-resourced corporation will clear the bar easily while a self-represented tenant with an AI-transcribed voicemail may not.

Honest mistakes. Deepfakes are the headline, but plenty of AI problems are accidental: a bad prompt, a hallucinated detail, an auto-enhanced photo nobody thought twice about. The OBA made this point too. The rules should catch sloppy AI, not only sinister AI.

Access to justice. The Law Commission of Ontario urged that courts consider access to justice when applying these rules, and that the rules be reviewed in two or three years. This technology moves faster than any rules committee, so a built-in check-up is not optional.

So what should you do now?

If you are a litigator, start asking clients early whether any document, photo, recording or analysis has passed through an AI tool. Keep originals, record what tool was used, and assume you will need to explain it. It is good evidence management, and soon it will be expected.

If you run a business, look at where AI already touches your records: meeting transcripts, customer-call summaries, enhanced images. If those ever end up in a dispute, you want a paper trail. (This is exactly the kind of work I described in my piece on the rise of the legal engineer.)

If you are anyone else, know that what you share as a joke today could be someone’s exhibit tomorrow. And if you are still wondering who owns the things AI makes for you, I have thoughts on that too.

Ontario is putting real procedural teeth on AI evidence, and I think it is right to. If the final text keeps the presumption against undisclosed AI and adds practical support for ordinary litigants, other provinces should copy it. The uncle in the WhatsApp group can keep his goat. The courtroom deserves better.

FAQ

Are Ontario’s AI evidence rules in force?

Not yet. The Civil Rules Committee has sent a revised set of draft rules to Ontario’s Attorney General, and updated text is expected in November 2026. Until the rules are finalized and approved, courts rely on existing evidence law, and lawyers must still certify that the authorities in their factums are authentic.

What happens if AI-generated evidence is not disclosed?

Under the latest draft, as reported by Law Times, evidence that was made with AI but not disclosed would be presumed inadmissible once discovered. That includes deepfakes. In other words, the starting point would be exclusion, so hiding AI involvement would be a very risky strategy for any party.

Will lawyers have to hand over their AI prompts?

Not automatically. The draft leaves prompt disclosure to the judge, because prompts can reveal litigation strategy and may be protected by solicitor-client privilege. A court can still order it in the right case, so lawyers should keep careful records of how AI tools were used on evidence.

---

This article reflects my personal views and is for general information only. It isn’t legal advice and doesn’t represent the views of any organization I am affiliated to. If you need advice on your specific situation, talk to a qualified lawyer in your jurisdiction.