AI for Law Firms in Canada: What Law Societies Say and the Fake-Case Lesson

The short answer

Canadian law societies have not banned generative AI. They treat it like any other tool or assistant: you stay responsible for competence, confidentiality, supervision and honest billing, and you must verify every authority before it reaches a court. The firms that got into trouble did not use AI too much. They filed what it produced without checking it.

Interest in AI for law firms is no longer theoretical. The Thomson Reuters Institute's 2026 AI in Professional Services Report found that 40% of professionals say their organisation now uses generative AI, up from 22% a year earlier, and two-thirds of corporate respondents want their outside firms to use it. The same report found only 18% of organisations track a return on that investment. This guide covers what the regulators have said, what four Canadian rulings teach, where AI actually helps a small or mid-size firm, and a first 90 days that keeps you out of the case reports.

What the Law Society of Ontario actually asks of you

In April 2024 the Law Society of Ontario's Futures Committee tabled a white paper on licensee use of generative AI, supported by a quick-start checklist, best-practice tips and a summary of professional obligations in its Technology Resource Centre. As Osler's summary of the LSO guidance notes, Ontario was the fifth Canadian law society to publish guidance, after Alberta, Manitoba, Saskatchewan and British Columbia. The approach is to apply existing rules, not write new ones:

• Competence. Some tasks need your own judgment and cannot be handed to a model.
• Confidentiality. Client information must stay protected wherever it might be typed into a tool, so the security and data-use terms of that tool matter.
• Supervision. AI output is treated much like a non-lawyer's work product. You review it before it goes anywhere.
• Fees. Bill only for time actually spent, even when AI makes the work faster. Alternative fee arrangements are possible if they are fair and reasonable.
• Disclosure to clients. Not required in every case. The guidance lists factors: whether client information goes into the tool, whether the client expects you to do the work personally, and whether the use carries reputational risk.

British Columbia added commentary to its competence rule in 2024 saying lawyers should understand the benefits and risks of relevant technology, including risks to confidentiality, according to McCarthy Tétrault's review of Zhang v. Chen. If you practise outside Ontario, read your own society's guidance; the themes are consistent but the wording differs.

Four rulings every Canadian firm should know

Zhang v. Chen, 2024 BCSC 285. Counsel filed a notice of application citing two cases that ChatGPT had invented. The court declined special costs because there was no intent to deceive, but held the lawyer personally liable for ordinary costs of the application, ordered a review of her files, and said it would be prudent to tell the court and opposing parties when submissions include AI-generated content.

Federal Court notice, updated May 7, 2024. Documents prepared for litigation that contain AI-generated content must include a declaration in the first paragraph. Hicks Morley's summary of the updated notice explains that the declaration is needed where AI contributed content in a co-author-like way, not where it only suggested edits, and that "human beings must be involved to verify AI outputs."

Ko v. Li, 2025 ONSC 2965. A Toronto lawyer's factum cited cases that did not exist and one that said the opposite of what was argued. Justice Myers started a contempt process. It ended when she admitted the error, withdrew the factum and undertook at least six hours of professional development on ethics and technology, and agreed not to bill the client for the work, as set out in McCarthy Tétrault's analysis of Ko v. Li. The factum also lacked the certificate required by Rule 4.06.1(2.1) of Ontario's Rules of Civil Procedure, which asks counsel to confirm they are satisfied every cited authority is authentic.

Hussein v. Canada, 2025 FC 1060. In an immigration matter, counsel used an AI research tool, did not verify the citations, and disclosed the AI use only after four directions from the Court. The Associate Judge called AI "a perfectly valid tool for counsel to use" but found the late disclosure amounted to an attempt to mislead, and raised personal costs, according to McCarthy Tétrault's Federal Court summary.

The lesson from the fake-citation cases

Read the four decisions together and the pattern is plain. No court punished a lawyer for using AI. The problems were an authority nobody opened, a certificate nobody signed, and a disclosure that came too late. That gives you a simple rule for the firm: every case, statute and quotation in a filed document is pulled from a primary database and read by a person before it is cited, and the person who signs is the person who checked.

General chatbots are the riskiest place to start legal research because they generate plausible text rather than retrieving it. Tools built on retrieval from a defined set of documents reduce, but do not remove, that risk. Treat any citation as unverified until someone has opened it.

Where AI for law firms earns its keep

Task	Good fit?	Guardrail
Intake forms, conflict-check prep, appointment booking	Yes	No legal advice from the bot; route anything substantive to a person
Summarising your own transcripts, disclosure or long emails	Yes	Use an approved tool with no training on your data
First drafts of routine letters and precedents	Yes	Lawyer review before sending
Legal research and factums	With care	Every authority opened, read and certified
Public-facing chatbot answering legal questions	Rarely	The LSO flags the risk of inadvertent legal advice

The admin side is usually where a small firm sees time come back first. Intake, scheduling, reminders and document collection carry less professional risk and follow the same pattern as other professional services firms. If your practice is mainly immigration, our companion guide on AI for immigration consultants and legal-service businesses covers the CICC side.

A first 90 days for a small firm

1. Days 1 to 30: write the policy. List approved tools, banned uses and who signs off. Read the vendor's terms on data retention, model training and where data is stored. Our guide to connecting AI to business data safely lists the questions to ask.
2. Days 31 to 60: pilot two low-risk tasks. Intake and transcript summaries are typical. Track time saved and every error caught.
3. Days 61 to 90: add research with a verification step. Build the "open every authority" check into your filing checklist, add the Rule 4.06.1(2.1) certificate, and prepare a Federal Court declaration template. Decide your client-disclosure position using the LSO factors and note it in your retainer.

This is general information, not legal advice. Check your own law society's current guidance and any practice directions from the courts you appear in.

Key takeaways

• Law societies apply existing duties to AI: competence, confidentiality, supervision and fair billing.
• Courts have sanctioned unverified citations and late disclosure, not AI use itself.
• Ontario factums need a signed certificate on the authenticity of authorities; the Federal Court needs an AI declaration where AI generated content.
• Start with intake and summaries, then add research only with a mandatory verification step.

Frequently asked questions

Do Ontario lawyers have to tell clients they use AI?

Not in every case. The LSO guidance lists factors to weigh, including whether client information goes into the tool and whether the client expects the lawyer to do the work personally. You can set out your position in the retainer agreement.

Can a lawyer bill for time saved by AI?

The LSO guidance says to bill only for time actually spent. Fixed or alternative fees are possible if they are fair and reasonable, and a reasonable charge for the technology itself may be acceptable.

What happens if a court finds a fake AI citation in my materials?

Outcomes in Canada have included personal costs, a file review, a contempt show-cause process and mandatory professional development. Prompt admission and correction have mattered in how courts responded.

Sources

1. Law Society of Ontario publishes guidance for licensees with respect to the use of generative AI — Osler, Hoskin & Harcourt LLP, April 25, 2024
2. Landmark Decision about Hallucinated Legal Authorities in BC Signals Caution… But Leaves Questions About Requirement to Disclose Use of AI Tools — McCarthy Tétrault, March 21, 2024
3. Federal Court Issues Updated Notice Regarding the Use of AI in Court Proceedings — Hicks Morley, June 5, 2024
4. AI Legal Ethics: The Case of Ko v Li — McCarthy Tétrault, May 23, 2025
5. Verify and Disclose AI Use – Requirements from the Federal Court — McCarthy Tétrault, June 19, 2025
6. 2026 AI in Professional Services Report — Thomson Reuters Institute, 2026

How Flowmatix can help

We put this into practice for businesses in Toronto and around the world through our Legal Technology, ai agents & ai solutions, business process automation services. Want to know what it would take for your business? Get a personalised quote in about 3 minutes.