Boonstra v. City of Edmonton: Alberta Court Dismisses Appeal After Litigant Cites 13 ChatGPT-Hallucinated Cases — ‘Grave Affront to the Administration of Justice’ | Advocate Prakhar

Case at a Glance

Background

John Boonstra appealed a decision of the City of Edmonton’s Civil Service Local Authorities Appeal Board (CSLAC) — an employment-related administrative tribunal — to the Court of King’s Bench of Alberta. Boonstra appeared without legal representation. The City of Edmonton responded and, in reviewing Boonstra’s materials, identified at least 13 case citations that did not exist. What made this case particularly striking was the evidence of AI use: the pages of Boonstra’s submission containing these citations had visible ChatGPT footers — AI platform interface marks visible on printed output — making the use of ChatGPT unmistakable on the face of the documents.

The AI Issue

At the hearing, Boonstra acknowledged using ChatGPT and apologized to the court, saying he “got caught in the trap of taking AI at face value.” He asked the court to ignore the fictitious cases and confirmed he would not rely on them. The presiding judge explained the court’s policy on AI use — requiring a “human in the loop” who carefully and diligently reviews AI-generated content before it is placed before the court — and stated that submissions “underpinned by false case law inevitably cast serious doubt on the credibility and integrity of a litigant’s submissions as a whole.” The judge was not able to rely on any submission tied to the fictitious case law.

What the Court Decided

• Boonstra’s appeal was dismissed — the CSLAC decision was not plainly irrational and met the standards of fairness required by common law [merits dismissed].
• All submissions relying on the 13+ ChatGPT-hallucinated cases were excluded from the court’s consideration [exclusion of tainted submissions].
• The submission of false legal authorities to the court “constitutes a grave affront to the administration of justice, the seriousness of which cannot be overstated” [formal condemnation].
• The court articulated the “human in the loop” requirement for AI-assisted submissions — the AI itself is not the problem; the absence of human verification is [standard for permissible AI use in Canadian courts].
• Submissions underpinned by false case law cast doubt on the entire submission — not just the specific arguments the false cases were meant to support [credibility contamination].

“Submissions underpinned by false case law inevitably cast serious doubt on the credibility and integrity of a litigant’s submissions as a whole. The provision of false legal authorities to the Court constitutes a grave affront to the administration of justice, the seriousness of which cannot be overstated.”

— Court of King’s Bench of Alberta, Boonstra v. City of Edmonton, 2026 ABKB 308

The India Angle

Indian Law Equivalent

The Alberta court’s “human in the loop” requirement resonates with the accountability structure under Indian professional conduct rules. Section 34 of the Advocates Act, 1961 empowers High Courts to make rules regarding the practice of advocates — including requiring verification of authority cited. Several Indian High Courts have issued administrative circulars requiring advocates to certify the authenticity of citations; the Supreme Court’s e-filing portal requires specific fields identifying cited provisions. The Boonstra case reinforces why these verification requirements are not bureaucratic formalities but substantive safeguards.

Bar Council Rules

Bar Council of India Rule 22 (no false statements) and Rule 15 (fair conduct) apply. The Alberta court’s framing — that false case citations undermine the credibility of the entire submission — is a powerful tool for Indian advocates who encounter opponents using AI-generated research: a showing that even one citation is fabricated can legitimately invite the court to scrutinize the reliability of the entire opposing submission.

Practical Advice for Indian Advocates

• ChatGPT and other LLMs often produce visible interface artifacts when output is printed directly — review your submissions before filing to ensure no platform UI elements (footers, watermarks, interface text) are visible in the document.
• When you discover a fabricated citation in an opponent’s filing, bring it specifically and immediately to the court’s attention — courts in Canada, the US, and increasingly India will treat this as going to the reliability of the opponent’s entire submission.
• The “human in the loop” standard articulated in Boonstra is the minimum Canadian standard — apply it proactively: every AI-generated draft must pass through a named responsible advocate who personally verifies each citation before filing.

Quick Takeaways

• ChatGPT interface footers visible on printed documents immediately prove AI use — check your prints before filing.
• One fake citation can undermine the credibility of your entire submission, not just that argument.
• Canada’s “human in the loop” standard is the international minimum for AI-assisted legal submissions.

Deep Dive: The Credibility Contamination Principle and the International AI Hallucination Standard

Boonstra v. City of Edmonton establishes what can be called the “credibility contamination principle” in Canadian law: when a litigant submits documents containing fabricated legal authority, the court cannot rely on any part of those submissions that is tied to the false citations. The practical effect is devastating — if 13 cases in your brief are fictitious, and those cases were used to support multiple arguments throughout the document, the court may find it impossible to disentangle which arguments were supported only by the fake cases and which stood independently. The safest judicial response — protecting the integrity of the court’s process — is to exclude the entire submission or assign it no weight.

The visibility of ChatGPT’s interface elements on Boonstra’s printed documents is an important practical lesson. When users access ChatGPT through a web browser and print the output, the printed document may include interface elements: the ChatGPT logo, session footers, the word “ChatGPT” in the header, or similar markers. Boonstra’s documents apparently had “footers” of this type — making it unmistakable to the City’s lawyers and to the court that the text had been directly printed from a ChatGPT session without any intermediate formatting or review. This is the opposite of a professional legal submission: it is raw AI output filed as legal argument, without even the minimal step of copying the content into a word processor and removing the interface elements.

The Alberta court’s “human in the loop” formulation has become an internationally influential standard. It appears in decisions from New Zealand, Canada, and is referenced approvingly by US courts. The principle is simple: AI can assist in drafting, research, and organization, but a qualified human must review every citation and every legal claim before the document is submitted to a court. The AI is a drafting tool; the advocate is the responsible professional. This division of responsibility is not new — it is the same principle that governs the use of paralegals, junior associates, and legal research services. The advocate who files the document is personally responsible for its accuracy, regardless of who or what produced the first draft.

For Indian advocates — particularly those practicing in civil and labour matters where self-represented litigants are common — the Boonstra case has an important practical implication. When you face a self-represented opponent whose submissions contain plausible-looking but unveririfable case citations, check those citations. If you find fabrications, bringing them to the court’s attention is not just a tactical move — it protects the court’s time and integrity, and the court’s response (as in Boonstra) may be to discount the opponent’s entire submission. This is a legitimate, ethical, and powerful litigation tool in the AI era.