H.B. v. Conseil des Écoles Publiques de l’Est de l’Ontario: Canadian Tribunal Orders Digital Case Copies After AI-Fabricated Citations | Advocate Prakhar

Case at a Glance

Background

H.B., a minor represented by litigation guardian O.B., filed a human rights application before the Human Rights Tribunal of Ontario against the Conseil des écoles publiques de l’Est de l’Ontario (a French-language public school board), as well as two individual school officials. The application was scheduled for a four-day merits hearing, and the matter proceeded through several rounds of pre-hearing motions (RFOPs — Requests for Orders during Proceedings).

In RFOP1, the applicant sought an order compelling document production, among other procedural relief. In support of the argument that schools must document and produce certain records, the applicant cited “Shah v. Toronto DSB, 2021 HRTO 342” for that proposition. The adjudicator searched CanLII and found that the citation “2021 HRTO 342” in fact refers to a case called Awad v. Petrakos — an application that was dismissed for reasons unrelated to the proposition being argued. The adjudicator concluded this was AI-generated fake case law and characterised it as imposing unnecessary burdens on the respondents and the tribunal.

The AI Issue

The pattern here is instructive in its specificity. The AI did not entirely invent the case citation — it generated a plausible-sounding HRTO citation number (2021 HRTO 342) and a plausible-sounding case name (Shah v. Toronto DSB). In reality, the citation number belongs to a different case. This “mismatched” hallucination — real citation number, wrong case name — is actually more dangerous than a fully invented citation, because superficial searching might find the citation number and assume the case is correct without verifying the name. Only a thorough check (here, using CanLII to look up the actual case at that citation) reveals the mismatch.

What the Court Decided

• RFOP1 dismissed: the applicant failed to demonstrate that required documents had been withheld, and the AI-fabricated case citation undermined the legal argument.
• RFOP2 and RFOP4 granted in part: applicant permitted to rely on Ottawa Police Service report and documents obtained through access to information requests; publicly available documents excluded.
• RFOP3 (adjournment request) partially granted: hearing dates of May 12 and 13, 2026 rescheduled; May 19 and 20 dates to proceed as scheduled.
• Key remediation order: applicant required to provide digital copies of all case law they intend to rely upon in any future proceedings before the tribunal, with relevant sections highlighted for ease of reference.
• Adjudicator warned the applicant against further “abuse of process by reliance on fake case law and Rules” in a direct closing paragraph of the order.

“In my view this represents the use of artificial intelligence to rely on fake case law. The reference to non-existent cases to miscite propositions that do not exist imposes [unnecessary burdens on the respondents and the tribunal].”

— Vice-Chair Robert Patchett, Human Rights Tribunal of Ontario, April 14, 2026

The India Angle

Indian Law Equivalent

India has quasi-judicial human rights bodies including the National Human Rights Commission (NHRC) and State Human Rights Commissions under the Protection of Human Rights Act, 1993. These bodies exercise procedural powers analogous to those of the Ontario HRTO. The NHRC’s procedural regulations empower it to dismiss complaints based on misleading or false information and to take action against parties who misrepresent the law or facts before it. While there is no specific Indian precedent on AI-hallucinated citations before human rights commissions, the general principle — that parties who cite non-existent authorities before quasi-judicial bodies abuse the process — is embedded in the commissions’ constitutive statutes.

Bar Council Rules

Advocates appearing before quasi-judicial bodies in India are subject to the same duty of candour as they are in courts. BCI Rule 49 (no misstatement of law) and Rule 52 (accurate citation) apply to proceedings before the NHRC, State Human Rights Commissions, Labour Courts, Administrative Tribunals, and similar bodies. The verification obligation is not relaxed for administrative tribunals; if anything, the informality of some tribunal proceedings may mean that a fabricated citation will go undetected longer — but when detected, the conduct is treated as more dishonest, because the party exploited a less-scrutinised forum.

Practical Advice for Indian Advocates

• When citing administrative tribunal decisions (e.g., NHRC, Labour Court, CAT), verify citations using official tribunal websites or authenticated databases such as Manupatra — CanLII’s role in Ontario is played by different databases in India, but the verification obligation is identical.
• The “mismatched” hallucination pattern — correct citation number but wrong case name — is particularly insidious: always verify both the citation coordinates and the case name, and read the relevant passage before filing.
• The digital-copy-with-highlights requirement imposed in this case is a model for what tribunals may order in India: if you represent a client before a tribunal that has concerns about citation accuracy, proactively offer to provide verified digital copies of every authority — it demonstrates good faith and pre-empts a compulsory order.

Quick Takeaways

• AI-hallucinated citations can involve mismatched citation coordinates — a real citation number leading to a different case — not just fully invented citations; verify both the number and the case name.
• Administrative tribunals treat AI-fabricated citations as abuse of process that burdens both the respondent and the tribunal — the duty of accuracy extends beyond formal courts.
• Forward-looking verification requirements (provide digital copies of all future authorities) are becoming a standard AI-hallucination remediation tool in common-law jurisdictions.

Deep Dive: AI-Hallucinated Citations in Administrative Tribunals — A Different but Equally Serious Problem

The H.B. v. Conseil des Écoles Publiques ruling is significant because it extends AI-hallucination jurisprudence from formal courts to administrative tribunals. This is an important development. Administrative tribunals across common-law jurisdictions handle a vast volume of disputes — employment, human rights, immigration, social welfare, professional licensing — and they are increasingly being used by self-represented parties. AI tools lower the barrier to filing procedural motions and supporting submissions in these proceedings, creating a new vector for fabricated citations in forums that may lack the institutional resources to check every citation.

Ontario’s HRTO adjudicator used CanLII, Canada’s free legal database, to check the applicant’s citation. This took seconds and revealed the fabrication immediately. The ease of detection means that the failure to verify citations before filing is indefensible — it is not a time-pressure issue or a resource issue. It is a choice not to verify. Tribunals, once they recognise this, are understandably intolerant.

The remediation measure adopted here — requiring digital copies of all future case authorities with relevant passages highlighted — is particularly well-suited to tribunal proceedings. It creates an ongoing verification obligation: the party cannot cite a case without simultaneously providing the tribunal with the means to check the citation’s accuracy. This is a form of structural transparency that requires no additional verification staff on the tribunal side. It also aligns with the practical reality that self-represented parties using AI need to be nudged into the habit of actually opening and reading the cases they cite.

For Indian advocates appearing before administrative tribunals, the lesson is prospective: implement the digital-copy-with-highlights practice as a self-imposed standard for every tribunal submission, not just when ordered to do so. It signals professional seriousness, aids the adjudicator’s preparation, and creates a personal discipline that prevents AI-hallucinated citations from slipping through. The small additional effort of preparing highlighted PDFs of cited authorities is trivial compared to the professional risk of a citation hallucination being detected in a quasi-judicial proceeding.