What happened
Queensland Law Society’s Proctor reports that Federal Court Justice Michael Lee ordered parties in a Fair Work dispute to confer with an expert about using AI to prepare evidence for mediation and trial. The order concerns a representative proceeding in the Federal Court’s Victorian registry involving allegations that McDonald’s failed to pay managers for work performed before and after shifts.
Queensland Law Society’s Proctor reports that Justice Michael Lee made the order under section 37P(2) of the Federal Court of Australia Act 1976. The report says the order requires the parties to investigate whether AI could help collate and analyze data, prepare chronologies and document summaries, and manage and interpret documents. The parties must then provide the court with a report and/or draft protocol describing the results.
The order arises from a Fair Work dispute involving McDonald’s and allegations that managers were not paid for work performed before and after shifts. According to Proctor, the case requires the parties to understand the number and circumstances of represented group members, their employment classifications, their full-time, part-time or casual status, relevant shifts in business records, and remuneration information. The report describes the proceeding as a representative action in the Federal Court’s Victorian registry.
Proctor reports that Justice Lee’s 20-page reasons were delivered on 28 August. In those reasons, he wrote that generative AI could advance the civil procedure system’s goals by increasing efficiency and reducing legal costs. His reasoning, as reported, focused less on whether a machine could draft a competent submission and more on whether technology could reorganize litigation processes that developed when humans had limited capacity to collect, relate and synthesize information.
The judge’s stated aim, according to Proctor, is to use technology for work that does not itself require human judgment while concentrating professional and judicial attention on matters that do. Proctor reports that the order does not require the parties to become software developers, does not prescribe a provider or technology, and does not settle the eventual protocol. The supplied source refers readers to the case, but does not reproduce the judgment; the exact order and any later developments are therefore not independently confirmed by the supplied material.
Source details: qlsproctor.com.au ↗
Why it matters
The order places AI at the center of formal litigation case management rather than treating it only as a drafting or search tool. It could influence how courts and lawyers handle large bodies of dispersed records while preserving human responsibility for judgment, privilege and legal accuracy.
The reported order is consequential because it treats AI as a possible component of the evidence-management process in an active court proceeding. That is different from using a chatbot to draft correspondence or search for authorities. The proposed tasks—linking records to people and periods, producing chronologies, summarizing documents and interpreting large collections—can shape what lawyers and judges see as the important facts of a case.
The practical problem described by Justice Lee is the scale and fragmentation of litigation records. Proctor reports that the relevant information exists but is spread across business records held in different repositories and formats. Turning that material into an intelligible account can require substantial manual labor. In a representative proceeding, the judge said, unnecessary process and expense may be multiplied across many represented people.
The approach also assigns clear limits to automation. Proctor reports that Justice Lee said the court should not entrust subjective adjudicative tasks to a machine. Any process would need to preserve traceability to primary records, allow outputs to be tested and corrected to some level, protect privileged and confidential material where necessary, and remain subject to appropriate human supervision. Those conditions address the central risk that a convenient summary could obscure missing, misclassified or incorrectly interpreted evidence.
The report does not establish that AI will reduce costs, improve accuracy or produce a usable litigation protocol. It describes an investigation ordered by the court, not a completed deployment or a validated system. The public significance is therefore prospective: the proceeding may provide a concrete test of whether AI can assist with high-volume evidence work without weakening accountability, reviewability or the parties’ ability to challenge the underlying material.
What to watch next
The parties must investigate possible uses of AI and provide the court with a report and/or draft protocol by 6 November 2026. Important unknowns include which technology, if any, will be used; what testing will show; how errors will be corrected; and whether the court approves a repeatable process for future proceedings.
The immediate milestone is 6 November 2026, when Proctor reports that the parties must provide the court with their findings and/or a draft protocol. That document should clarify whether the parties identified suitable uses, what expert input they obtained, and what technical and legal controls they consider necessary. The supplied source does not say whether the court has already approved any particular tool or whether the parties have begun using one.
Traceability will be a key issue. A credible process would need to connect summaries, classifications and analytical outputs back to the underlying business records so lawyers and opposing parties can inspect and challenge them. The judge’s reported emphasis on testing and correction also leaves an important question: what level of error is acceptable for each task, and who is responsible for detecting and fixing mistakes before material is used in mediation or trial?
Confidentiality and privilege will require close attention. The source says the process must protect privileged and confidential material where necessary, but it does not explain whether data could leave a party’s controlled environment, whether a provider would retain inputs, or how access would be logged. It also does not identify the model, vendor, security design, evaluation dataset or review standard that might ultimately be adopted.
The broader outcome to watch is whether the court treats the experiment as a narrow case-management measure or as a model for other complex proceedings. Proctor reports that Justice Lee did not prescribe technology or determine the eventual protocol, so the parties’ report and the court’s response will matter more than the initial order. No claim can yet be made about actual efficiency gains, accuracy, cost savings or changes to Australian court practice.