AI Compliance7 min read

Canada's First Criminal AI Contempt: What Ontario Lawyers Must Document in 2026

A Toronto lawyer faced criminal contempt for lying about an AI error. Ontario’s three regulatory systems now require every lawyer to document AI compliance.

What You'll Learn

The Ko v. Li case established that AI errors become criminal liability when lawyers lie to cover them — not when the tool fails. This article maps the three overlapping compliance systems Ontario lawyers now need to document: LSO AI guidance, Legal Aid Ontario's annual self-report requirement, and Bill C-36.

The Case That Changed the Stakes

The Ko v. Li decision turns on what happened after an AI error, not on the error itself.

In May 2025, Toronto lawyer Jisuh Lee of ML Lawyers filed a factum containing non-existent case law generated by ChatGPT (Ko v. Li, 2025 ONSC 2965, Ontario Superior Court). When Justice Myers confronted her in open court about whether she had used ChatGPT, she denied it. She said: "I did not."

On December 4, 2025, Justice Myers released a 15-page endorsement referring Lee to the Attorney General of Ontario for criminal contempt of court (Ko v. Li, 2025 ONSC 6785, CanLII). Justice Myers referred Lee for criminal contempt because she lied to him in open court. The hallucinated citations alone would have resulted in a cost order.

In a separate Ontario matter, Law Times reported that a lawyer who submitted AI-fabricated citations was hit with $31,150 in costs to the Law Society of Ontario (Law Times, 2025).

💡

Two Ontario cases in 2025 established a cost floor for AI citation errors and escalated one to criminal contempt proceedings. The factor that determines which category: what the lawyer said afterward.

Why Three Regulatory Systems Now Overlap

Most lawyers read the Jisuh Lee case as a professional discipline story. The full picture adds two more layers, and they reinforce each other.

LSO Professional Obligations

The Law Society of Ontario's AI guidance applies under two existing rules: Rule 3.1-2 (competence) and Rule 3.3-1 (confidentiality) (Augure AI, LSO AI Guidance for Ontario Lawyers). Lawyers must understand what their AI tools do, verify every output before relying on it, and protect client information from entering AI systems with inadequate data governance.

Both rules predated generative AI by decades. They apply to ChatGPT the same way they apply to any other tool a lawyer uses to draft client materials or court filings.

Legal Aid Ontario's Mandatory Self-Report

Starting January 2026, every roster lawyer doing Legal Aid work must annually confirm compliance with LSO materials on generative AI (Legal Aid Ontario, December 2025). LAO will decline payment for AI-generated content containing inaccuracies. Lawyers who miss the March 31 filing deadline lose access to LAO systems until they file.

This is an annual documented compliance obligation — on paper, with a signature.

Bill C-36 (PPCDA)

Bill C-36, the Protecting Privacy and Consumer Data Act, was introduced in Parliament on June 15, 2026 (Parliament of Canada). When enacted, it will apply to every organization collecting or using personal information in commercial activities — including law firms. The PPCDA requires documented consent processes when AI systems make recommendations with "legal or similarly significant" effects on individuals. Penalties for serious violations reach $25M CAD or 5% of global revenue (IAPP analysis, 2026).

Ontario law firms running Clio Duo, Harvey AI, Spellbook, or general-purpose AI for client matters need consent language and data governance documentation that most have not yet prepared.

Not sure where AI fits in your operations?

Take the Free AI Readiness Assessment

What Compliance Actually Requires

The practical obligation across all three layers reduces to three documented items.

A Written AI Policy

The LSO guidance requires lawyers to demonstrate they understand their tools and have implemented supervision protocols. The LAO self-report requires annual signed confirmation of LSO compliance. Both are satisfied by a written AI policy that names: which tools are approved for which use cases, what client data categories can be entered, and how outputs are reviewed before court or client use.

Neither requirement is satisfied by a verbal agreement or a shared understanding — both require a dated written document.

📊
Example

A 4-person litigation firm documents that ChatGPT-4o is approved for drafting research summaries, that no client names or matter details may be entered, and that all case citations must be verified against CanLII before inclusion in any court filing. The policy takes 90 minutes to draft and satisfies the LSO guidance documentation requirement and the LAO self-report for all roster lawyers at the firm.

Updated Retainer Language for AI-Assisted Work

Bill C-36 requires organizations to explain, on request, how automated systems produce recommendations affecting individuals. Law firms that use AI to draft client advice, strategy documents, or correspondence need consent language in their retainer agreements — most of which were last updated before generative AI was in widespread use.

This is a standard retainer amendment. But firms that have not reviewed their retainers since 2023 are operating without coverage they will need.

A Verification Protocol for Court Filings

The Ko v. Li case is specific about what created the original error: Lee did not verify the AI output before submitting it to court. The citations were fictional. A mandatory verification step — checking every case citation against a reliable database before submission — prevents this category of error.

The criminal contempt proceedings arose from the cover-up, not the error. Courts have shown they treat prompt, voluntary disclosure differently than denial. A verified output that later turns out to be wrong, disclosed immediately, enters professional conduct territory. A lie told in open court enters a different territory entirely.

Result

A litigation lawyer implements a mandatory CanLII check for all AI-generated case citations before any court submission. After three months, zero citation errors reach court. The check adds approximately 20 minutes per filing. The LAO self-report was filed on time with documentation of the verification protocol in place.

💡

Voluntary, immediate disclosure after an AI error moves the matter into professional conduct. Denial to a judge in open court moves it into criminal territory.

The Forward-Looking Pressure

Three enforcement mechanisms are now active simultaneously. Courts are alert to AI-generated content and have shown they will ask directly. LSO enforcement will have a documentary record — the LAO self-report — showing which lawyers claimed compliance and when they filed. Bill C-36 adds a third reporting architecture when enacted, with penalty levels calibrated to GDPR scale.

67 to 72% of Ontario solo and small-firm lawyers now use AI tools, but only 18% obtain client consent before doing so (LawNext, 8am AI in Legal Report, March 2026). The compliance infrastructure has not kept pace with adoption. Firms that have documented their AI practices — written policy, updated retainers, verified filings — have satisfied all three regulatory layers with roughly five to ten hours of work. Firms that have not are carrying undocumented exposure across three overlapping systems at once.

If your firm has not reviewed its AI practices since 2023, our AI Readiness Assessment identifies exactly where the gaps are and produces a written report your firm can use directly for LSO compliance documentation.

💡
Key Takeaways
  • Criminal contempt proceedings in Ko v. Li arose from lying to a judge, not from the AI error itself
  • Legal Aid Ontario requires an annual signed self-report on LSO AI compliance from all roster lawyers starting January 2026
  • Bill C-36 will require documented consent processes for AI tools affecting clients' legal positions when enacted
  • Three documents cover the practical obligation: a written AI policy, updated retainer language, and a verification protocol for court filings
  • 67-72% of Ontario solo and small-firm lawyers use AI, but only 18% obtain client consent — the documentation gap is the risk

Related Articles

Frequently Asked Questions

What happened in Ko v. Li, 2025 ONSC 6785?
Toronto lawyer Jisuh Lee of ML Lawyers submitted AI-hallucinated case citations in a factum, then denied using ChatGPT when questioned directly by Justice Myers in open court. Justice Myers referred her to the Attorney General of Ontario for criminal contempt of court in December 2025 — making this the first Canadian case where AI misuse escalated from professional negligence to potential criminal proceedings. The referral was triggered by the deliberate misrepresentation, not the AI error itself.
What does the Law Society of Ontario require from lawyers using AI?
The LSO's AI guidance applies under Rules 3.1-2 (competence) and 3.3-1 (confidentiality). Ontario lawyers must understand their AI tools' limitations, verify all outputs before relying on them in court or client work, and protect client confidential information from entering AI systems without adequate data governance. Legal Aid Ontario additionally requires annual self-report confirmation of LSO AI compliance from all roster lawyers starting January 2026.
How does Bill C-36 (PPCDA) affect Ontario law firms?
Bill C-36, the Protecting Privacy and Consumer Data Act, was introduced June 15, 2026 and will apply to any organization handling personal information in commercial activities — including law firms. It requires documented processes when AI systems make recommendations with 'legal or similarly significant' effects on individuals. Penalties for serious violations reach $25M CAD or 5% of global revenue. Law firms running Clio Duo, Harvey AI, Spellbook, or general-purpose AI for client work need to document consent and data governance practices under PPCDA on top of existing LSO obligations.
What three documents reduce AI compliance risk for Ontario lawyers?
Three items cover the practical obligation across LSO guidance, LAO self-report, and Bill C-36: (1) A written AI policy naming approved tools, approved data categories, and how outputs are reviewed before court or client use. (2) Updated retainer language covering client consent for AI-assisted work where AI outputs affect the client's legal position. (3) A verification protocol for court filings requiring every AI-generated case citation to be checked against CanLII or another reliable database before submission. The Ko v. Li case shows that prompt disclosure of errors is protective — lying is not.