What Ontario’s authenticity certificate does not certify

Pouya Shafabakhsh
Aug 24
6 min read

Law360 Canada (August 17, 2026, 9:14 AM EDT)

The Law Society Tribunal’s six-month suspension in Law Society of Ontario v. Lee, 2026 ONLSTH 136 turned in part on a factum containing four defective authorities. Three of them did not exist.

The fourth one did.

This article was originally published by Law360 Canada (LexisNexis Canada Inc.) on August 17, 2026. Read the original article on Law360 Canada.

What Ontario’s authenticity certificate does not certify

Authentic, and wrong

Ontario already had a rule aimed at this problem. Since Dec. 1, 2024, O. Reg. 384/24 has required every factum under the Rules of Civil Procedure to be accompanied by a statement from the filing lawyer, or someone authorized on the lawyer’s behalf, certifying satisfaction as to the authenticity of every authority cited. That is rule 4.06.1(2.1). The Court of Appeal’s own summary of the certification requirement, set out under the heading “Changes Regarding Civil Appeal Factum Requirements” on the Civil Rules Committee page of the Court of Appeal for Ontario’s website, explains that the requirement responded in part to reported North American incidents of counsel citing authorities that were not authentic but had been made up by artificial intelligence.

Rule 4.06.1(2.2) then makes the certificate workable in practice. Authorities published on a government website or by a government printer, on a court’s website or by a commercial publisher of court decisions are among the sources the rule presumes authentic, absent evidence to the contrary. Without that presumption, certifying a factum carrying a long list of authorities would be unmanageable.

Now run the fourth citation through the regime. The decision is real. If it appeared in any of the sources listed in rule 4.06.1(2.2), it was presumed authentic, and the presumption was correct, because the case genuinely exists. Counsel could sign the certificate in complete good faith and be right.

In my reading, that is the structural point the certification reform leaves open. The certificate addresses whether an authority is genuine. It does not address whether the factum has described what that authority holds. A citator carries the same boundary, reporting treatment history rather than confirming that a summary matches a judgment. Fabricated authorities now meet two independent controls. A genuine authority carrying a misdescribed holding meets neither.

Three questions, usually answered as one

Does the authority exist? Is it still good law? Does it say what the factum says?

Most verification routines answer the first two and assume the third. That assumption held for as long as whoever cited a case had ordinarily read it. Generative drafting broke it quietly, and no better database will repair it, because the failure is a reading failure rather than a retrieval failure.

The operational consequence is narrow and worth stating plainly. Existence and currency can be delegated to a clerk or a database. Content cannot, at least not for the authority an argument actually turns on. Someone has to open the judgment and read the paragraph.

The signature carries it

It makes no difference who produced the text: a law clerk running research at your direction, an associate drafting the argument, an outsourced research service, or you at eleven at night with a subscription and a deadline in the morning.

Rule 6.1-1 of the Rules of Professional Conduct makes a lawyer responsible for directly supervising non-lawyers to whom tasks and functions are assigned. Knowing that a tool was used and not checking its output is a supervision failure. Not knowing a tool was used is also a supervision failure, because the rule asks what a lawyer had in place, not what a lawyer happened to notice. There is no reading of it under which responsibility settles somewhere further down the corridor.

Two obligations ran in parallel in Lee. Rule 3.1-2 requires a lawyer to perform any legal services undertaken on a client’s behalf to the standard of a competent lawyer, and Rule 3.1-1 defines that lawyer as one who implements the chosen course of action through appropriate skills, including legal research. The commentary to Rule 3.1 already addresses technological competence, so nothing here required a new rule. The account given after the citations were raised engaged Rule 5.1-2, which prohibits counsel from knowingly attempting to deceive a tribunal by offering false evidence or misstating facts or law.

The candour and confidentiality provisions are easy to conflate, and the numbering does not help: Rule 3.3-1 is confidentiality, not candour. The distinction matters operationally. A firm that builds its AI training on 3.3-1 will teach its people to guard client data but say nothing about what to do when a judge asks where a case came from.

The tribunal itself was careful about the boundary. It said that using an AI tool to help prepare court submissions is not, by itself, professional misconduct. In a LinkedIn post on her profile page, Professor Amy Salyzyn, among others, has cautioned against reading the sanction as a tariff for AI misuse, and that caution is sound. The dishonesty finding makes the outcome difficult to generalize. The verification gap sitting underneath it generalizes without difficulty.

What is actually at risk

The visible consequences of a matter like this are the suspension and the costs order. They are not the only ones. A published discipline decision is permanent and searchable. It surfaces on a name search by a prospective client, a referring firm or opposing counsel years afterward, stripped of the mitigating context the panel had in front of it. Separately, an authority described inaccurately in a filed factum has a short causal chain to an adverse result for a client. That is why Ontario practitioners are better served knowing in advance what LAWPRO treats as a reporting trigger, and how AI-assisted work product is handled under their coverage. Those are questions to ask before a filing rather than during a claim.

Two regimes, one duty

Counsel practising on both sides of the border now work under rules that diverge on disclosure and converge on verification. New York’s Part 161 of the Rules of the Chief Administrator, in force since June 1, 2026, permits the use of AI tools in preparing papers and does not require disclosure of that use. The model rule it supplies places the obligation on whoever signs to review the paper and independently ensure that it contains no fabricated or fictitious cases, statutes or other material. Individual judges may adopt a stricter part rule.

Ontario requires a certificate and no disclosure of tool use. New York requires no certificate and no disclosure, but states the verification duty directly. Different instruments, the same underlying expectation: the person who signs has satisfied themselves that the information is correct. Neither regime treats a checklist, a detector or a vendor assurance as self-proving.

What actually closes the gap

Two responses cost nothing. The first is establishing which tools are genuinely in use across a practice, on which matters, and what client material has passed through them. The pattern security teams once called shadow IT now runs on generative tools, and adoption outpaces policy in most organizations. A questionnaire and a candid conversation will surface most of it.

The second is treating verification as a discrete step before a document leaves the office, with a named reviewer and a short written record: which passages had tool assistance, where each authority was retrieved from, and whether the pinpoint passage supports the proposition it is offered for. Recording it converts diligence from something asserted later into something demonstrated. For a sole practitioner, it means a reciprocal arrangement with a colleague. That is a structural point about who has a second reader, not a comment on anyone’s capability. Self-review under deadline is the condition the problem arises from, and it arises wherever the second reader is missing.

One response costs money. Consumer terms of service generally permit retention and human review of submitted content, which sits awkwardly beside a privilege claim a firm would want to defend. Keeping client material out of consumer tools answers that for a small practice. For a firm carrying sensitive litigation at volume, an isolated environment with no outbound connection to third-party infrastructure is what data sovereignty means in operation rather than in a policy binder.

The four-minute test

Pull a factum your office filed this month. Take the authority the argument leans on hardest. Open the judgment, find the pinpoint and read the paragraph on either side of it.

If that takes four minutes, the control is affordable at any firm size. If nobody in the office can say who did it last time, that is the finding, and it says nothing about any one lawyer.

Pouya Shafabakhsh is a principal judicial forensic AI auditor at Radsam Academy of AI Sovereign Governance (Radsam Trade Inc.), and the author of North America’s Litigation AI Governance, Risk and Compliance (GRC) and Judicial Forensic AI Audit Methodologies. He is not a lawyer and does not provide legal advice.

The opinions expressed are those of the author and do not reflect the views of the author’s firm, its clients, Law360 Canada, LexisNexis Canada or any of its or their respective affiliates. This article is for general information purposes and is not intended to be and should not be taken as legal advice.

Interested in writing for us? To learn more about how you can add your voice to Law360 Canada, contact Analysis Editor Yvette Trancoso at Yvette.Trancoso-barrett@lexisnexis.ca or call 905-415-5811.

It was a real decision. On the tribunal’s findings as reported, the tool had inverted its holding, describing an estate trustee as removed when the court had dismissed those claims. Ontario litigators should be looking hardest at that citation, because it is the one their existing controls were never built to stop.