Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence within the justice system. It supports my work in teaching, lecturing, and writing about AI and the law and is published to promote my practice. Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI Law.
1. Introduction
An occupational therapist from Far North Queensland has taken on her regulator and AI Hallucinations have taken centre stage. In LJY v Occupational Therapy Board of Australia [2025] QCAT 96 the practitioner’s written submissions quoted a completely fictional Court of Appeal decision confidently generated by ChatGPT. Deputy President Dann not only spotted the phantom authority but also appended Queensland’s Guidelines for Responsible Use of Generative AI by Non-Lawyers and delivered a concise tutorial on large-language-model fallibility.
Here, I will focus on the case and then explain how there appears to be a divergence in judicial philosophy between Australia (Queensland) and the Courts of England and Wales.
2. The Facts of the Case and Where AI Featured
The Board placed public‑facing conditions on the applicant’s registration after a parent complained about her NDIS functional‑capacity report. The conditions required costly one‑to‑one education and mentoring within six months, during the Christmas lull and were published on the AHPRA register. The applicant sought a stay under s 22(3) QCAT Act while she appeals the merits of the decision (unsatisfactory practice under s 178 National Law). The Tribunal granted the stay and, on its own motion, issued a non‑publication order shielding the child, family and practitioner.
The judgment’s most arresting section, to me at least, is an excursus on generative AI. The applicant’s written submissions cited a non‑existent Court of Appeal authority, allegedly sourced from ChatGPT. The Judge stated:
“Again, Ms LJY advances no evidence in support of this proposition. The Tribunal has been unable to locate Crime and Misconduct Commission v Chapman [2007] QCA 283, a decision Ms LJY has referred to in support of this part of her submissions.22 As the Tribunal can inform itself in any way it considers appropriate,23 I checked what ChatGPT had to say, if anything, about Crime and Misconduct Commission v Chapman [2007] QCA 283. ChatGPT told me broadly:…”
ChatGPT then obligingly produced a full‑blown summary: it claimed the case was a Queensland Court of Appeal decision on whether to stay a solicitor’s suspension, outlined putative facts, and even described the alleged parties’ arguments. Judge Dann records the chatbot’s output point‑by‑point, then explains “the case simply does not exist”.
The Judge continued:
“[26] It is important that Ms LJY, and other litigants before the Tribunal, understand that including non-existent information in submissions or other material filed in the Tribunal weakens their arguments. It raises issues about whether their submission can be considered as accurate and reliable. It may cause the Tribunal to be less trusting of other submissions which they make. It wastes the time for Tribunal members in checking and addressing these hallucinations. It causes a significant waste of public resources.”
Judge Dann responded with a 10‑paragraph tutorial on large‑language‑model fallibility and appended Queensland’s Guidelines for Responsible Use of Generative AI by Non-Lawyers . This is well worth a read, but I will not set it out here.
3. Comment
In every judgment I encounter (or receive), the hottest AI issue is hallucinated authorities. I will shortly post a South African decision on the same theme, and several U.S. cases are waiting in my queue.
A striking divergence seems to be emerging between England & Wales and Australia on how courts respond to AI use by self-represented parties. The Queensland Guidelines for Responsible Use of Generative AI by Non-Lawyers speak directly to court users. By contrast, the refreshed Artificial Intelligence (AI) Judicial Guidance for England & Wales is addressed to Judicial Office Holders. Nonetheless, their approaches differ in instructive ways.
Responsibility for citations
In Queensland, the litigant must verify every AI-generated citation before filing, because chatbots “may be inaccurate, incomplete, or out of date”. The Guidelines remind users that AI “is not a substitute for a qualified lawyer”. However, in E&W, judges are encouraged to spot AI-drafted material and to ask the litigant what accuracy checks have been performed. While the bench should warn that responsibility rests with the party, the initial policing impulse lies with the judge.
AI as the litigant’s sole legal adviser
The E&W guidance acknowledges that chatbots “may be the only source of advice or assistance some litigants receive” and that many lack the skills to spot errors. However, Queensland’s message is more categorical: AI cannot replace a lawyer and should never be relied on as the sole information source.
Given this distinction, a Judge in E&W might take a different approach to the applicant in LJY v Occupational Therapy Board of Australia when confronted with AI-generated material. We will have to see.
In later posts I will compare it these trends with other jurisdictions, especially the United States, but for now, I would welcome your thoughts, particularly from practitioners who have seen already handled AI-generated pleadings. Which approach feels more realistic in day-to-day casework, and how do we prevent the cure from becoming yet another barrier to entry?
