# HD-INC-011: The First Australian Lawyer Disciplined for AI Misuse Handed a Judge Fake Cases > A Victorian solicitor gave the Federal Circuit and Family Court a list of AI-generated authorities that could not be found. The regulator barred him from practising as a principal and required two years of supervised practice. - Source: https://useheadlights.com/library/HD-INC-011/ - Publisher: Headlights (useheadlights.com), author Ellie Harris - Published: 2026-06-07 - Company: Victorian Legal Services Board and Commissioner, Federal Circuit and Family Court of Australia (Division 2) - Sector: Legal services - Country: Australia - Year: 2024 - Behaviour in the record (Headlights reading): Automation bias; also Rules-behaviour gap - Governance domains: Human oversight - Licence: CC BY-NC 4.0. Cite as: Harris, E. (2026). The First Australian Lawyer Disciplined for AI Misuse Handed a Judge Fake Cases. Headlights AI Incident Library, HD-INC-011. https://useheadlights.com/library/HD-INC-011/ ## What happened It was reported that in 2024, an Australian solicitor appearing in the Federal Circuit and Family Court of Australia was asked by a judge to provide a list of previous court decisions supporting his client's position. The proceedings concerned a family law dispute between a married couple, in which he acted for the husband, and the request was an ordinary part of preparing legal arguments. The solicitor submitted a document containing case names, citations and summaries of the decisions he believed supported his client's case. When the judge attempted to read the authorities, however, she encountered a problem. The cases could not be located. They had been presented in the familiar format of legal citations, complete with references and explanations of what the courts had supposedly decided, but the authorities identified in the document did not exist as cited. The solicitor subsequently acknowledged that he had used AI-based legal research software to prepare the list and had not independently verified the references. He had relied on the information the software produced without checking the original judgments, apparently assuming that the cases and summaries were genuine. The incident was recorded in Dayal [2024] FedCFamC2F 1166, and the judge referred the matter to the Victorian Legal Services Board and Commissioner for consideration. The judge did not treat the incident as evidence that the solicitor was generally dishonest or incompetent. Her published reasons acknowledged the stress the matter had caused him and indicated that she did not consider the conduct likely to be repeated. Nevertheless, she considered the use of unverified AI-generated authorities sufficiently serious to warrant the attention of the professional regulator, particularly given the wider implications for lawyers using emerging technologies in court proceedings. On 19 August 2025, the Victorian Legal Services Board and Commissioner varied the solicitor's practising certificate, and it published a statement on the matter on 2 September. He was no longer entitled to practise as a principal lawyer, handle trust money or run his own practice, and could work only as an employee solicitor. He was also required to complete two years of supervised legal practice, during which he and his supervisor must each report to the Board every quarter. The decision was widely reported as the first public disciplinary action by an Australian legal regulator specifically involving a solicitor's misuse of AI. Although the technology was new, the professional responsibility at the centre of the case was not. Lawyers have always been expected to check the legal authorities they place before a court, regardless of how those authorities were found. ## What an auditable version would have shown The solicitor's admission established that he had used AI-based research software and had not independently checked whether the authorities were genuine. What the public record does not fully establish is how the software produced the references, what warnings accompanied its output or what records were retained during the research process. An auditable research process could have shown where each proposed authority came from and whether it had been located in a recognised legal database. This would have distinguished between a genuine judgment retrieved from a source such as AustLII and a reference generated by an AI model that had not yet been confirmed against an actual court decision. That distinction is important because fabricated legal authorities can be difficult to recognise by appearance alone. A citation may contain a plausible case name, court abbreviation, year and paragraph references, accompanied by a summary that sounds consistent with established legal principles. Unless the original decision is located and examined, the lawyer cannot know whether the case exists or whether it supports the argument being made. A CitationVerifier could assist by checking each reference against authoritative legal sources and identifying citations that could not be matched to genuine decisions. A VerificationGate could then require those discrepancies to be resolved before the document was submitted to the court, while a ConductRecord could preserve the checks undertaken and the person responsible for approving the final filing. The objective would not be to replace a solicitor's professional judgement with another automated system. It would be to make the verification process visible and consistent, ensuring that AI-generated research was treated as material requiring examination rather than as evidence that the research had already been completed. ## Where the gap was What makes this incident particularly revealing is the court's assessment of the solicitor himself. The judge accepted his apology and did not consider the conduct likely to be repeated. Yet the consequences were significant because the failure occurred at a point where the court was entitled to rely on the accuracy of the material being submitted. Headlights identifies automation bias in the solicitor's reliance on the AI research software. The tool produced references that looked credible, and he accepted them without independently confirming that the decisions existed. The apparent authority of the output was sufficient for the information to move from a research tool into a formal court document. The incident also illustrates a rules-behaviour gap between the professional obligations governing legal practice and the way the research was conducted. Solicitors are responsible for ensuring that the authorities they cite are genuine and accurately represented, but in this case the necessary verification did not take place before the document was filed. The problem was not the decision to use AI for legal research. Software can help practitioners identify relevant cases, organise material and explore possible arguments, provided its outputs are checked against reliable sources. The difficulty arises when the appearance of a completed piece of research is mistaken for evidence that the work has been properly verified. There is also a broader human tendency at work. When information arrives in a familiar professional format, people may be less inclined to question it, particularly when it appears to come from a tool designed for the task. A case citation with a convincing summary can look more trustworthy than an unsupported statement, even though both may have been generated without reference to a genuine judgment. This is why the verification step matters. It is not simply an administrative requirement or an additional precaution for people who use AI. It is the point at which a lawyer establishes that the information being relied upon is real, relevant and capable of supporting the argument placed before the court. These behavioural classifications are Headlights' analysis of the conduct described in the court and regulatory material, not additional findings made by the judge or the Victorian Legal Services Board and Commissioner. ## What governance should have looked like The most effective safeguard would have been to make independent citation checking a required part of preparing court documents. Before the list of authorities was filed, each reference should have been located in an authoritative legal source, the relevant judgment examined and the summary checked against what the court had actually decided. A CitationVerifier could assist with the initial process of identifying whether the cited decisions existed, but it would not remove the need for a solicitor to read and assess them. A genuine case can still be misquoted or relied upon for a proposition it does not support, so verification must extend beyond matching a citation to a database entry. A VerificationGate could require confirmation that the checks had been completed before a document progressed to filing. Where a reference could not be verified, the solicitor would need to correct or remove it rather than allowing the document to proceed with an unresolved citation. A ConductRecord could preserve the AI tools used during research, the authorities identified, the verification results and the approval given before submission. Such records could help firms demonstrate that appropriate checks had been carried out and assist in responding to questions from courts or regulators if concerns later arose. Australian courts have since developed guidance addressing the use of generative AI in legal proceedings, although requirements differ between jurisdictions and types of court material. The responsibility remains with practitioners to understand the rules applicable to their matters and to ensure that the documents they submit meet the court's standards. The lasting significance of Dayal is that the solicitor faced professional consequences for relying on material he had not checked, even though the judge did not consider the conduct likely to be repeated. The case illustrates how an apparently routine research shortcut can become a serious professional issue when the output of an AI system is accepted without independent verification. **Failure Pattern:** an Australian solicitor submitted a list of legal authorities generated with AI-based research software without checking whether the cited cases existed. The court identified the problem, referred the matter to the regulator and the solicitor's practising certificate was subsequently restricted. **Governance Principle:** AI-assisted legal research must be independently verified before material is submitted to a court. Firms should make citation checking a routine part of the filing process and retain records demonstrating how authorities were confirmed. The reference implementation of CitationVerifier, VerificationGate and ConductRecord is open source. It lives at [github.com/saffronandindia/headlights-oss](https://github.com/saffronandindia/headlights-oss), Apache 2.0 licensed and free to install. The repository is public now. ## Sources - [Statement on the 'Mr Dayal' matter (VLSB+C)](https://lsbc.vic.gov.au/news-updates/news/statement-mr-dayal-matter) - [Dayal [2024] FedCFamC2F 1166](https://www.austlii.edu.au/cgi-bin/viewdoc/au/cases/cth/FedCFamC2F/2024/1166.html) - [Australian lawyer caught submitting fake AI-generated cases allowed to keep practising under supervision (Law Gazette)](https://www.lawgazette.co.uk/news/australian-lawyer-who-submitted-fake-ai-cases-allowed-to-stay-in-practice/5124347.article) - [First Australian lawyer penalised for AI blunder (Information Age, ACS)](https://ia.acs.org.au/article/2025/first-australian-lawyer-penalised-for-ai-blunder.html) ## Related - [HD-INC-002](/library/HD-INC-002) - [HD-INC-003](/library/HD-INC-003) - [HD-INC-012](/library/HD-INC-012)