Federal Court Rules on AI Summary Costs in Insolvency Case

Judge orders vacation of hearing to enable verification of accuracy of summaries In a complex corporate insolvency proceeding where the plaintiffs repeatedly revised artificial intelligence (AI)-assisted summaries of thousands of documents and served them close to the scheduled hearing date, Australia’s Federal Court addressed the cost consequences of vacating the hearing. In Jahani v Qiu, in the matter of Ralan Property Services Pty Ltd (receivers and managers appointed) (in liq) (costs) [2026] FCA 1419, the corporate plaintiffs commenced a case concerning the alleged release and use of purchaser deposits within a group involved in over 30 property developments. The unpaid released deposits allegedly exceeded $288m when the group entered administration. The plaintiffs brought claims based on voidable transactions and accessorial liability. The proceeding involved evidence on valuation, industry practice, and solvency. On 11 August 2026, the plaintiffs served proposed summaries of more than 7,500 documents. Using generative AI or “super intelligence,” the plaintiffs extracted information from the documents. The plaintiffs’ solicitors then undertook a manual review and correction process. On 14 August 2026, the plaintiffs provided an updated version with the underlying documents. Upon spot-checking that material, the defendants found errors. The plaintiffs acknowledged these errors but challenged their materiality. The plaintiffs brought an application about the mode of proof of around 7,627 documents. An interlocutory hearing on 22 September 2026 addressed the plaintiffs’ s 50 application. The parties agreed that the s 50 mechanism would help avoid the tender of thousands of documents. Raising a concern regarding procedural fairness, the defendants sought to vacate the scheduled hearing. The defendants asserted that they did not have enough time and access to the source documents to verify the accuracy of the summaries served only weeks before the hearing, given their efforts to prepare for the upcoming trial. Upon considering the procedural history, the Federal Court adjourned the part-heard s 50 application, revised the timetable for the parties to identify and narrow their remaining differences, and relisted the hearing originally scheduled to start on 28 September 2026 to instead begin on 1 February 2027. The defendants requested a costs order against the plaintiffs based on [4.4] and [5.1] of the Federal Court’s Use of Generative Artificial Intelligence Practice Note (GPN-AI), which provided that presenting inaccurate AI-generated material might attract adverse cost consequences. The court disagreed that GPN-AI offered a distinct basis for a costs order against the plaintiffs. The court explained that GPN-AI did not aim to discourage the responsible use of AI tools to manage voluminous documentary material when accompanied by adequate disclosure and verification. In this case, the court saw no vice regarding responsibly using AI for a mechanical task involving voluminous documents. Citing Rogers v McDonald’s Australia Ltd (AI-use) [2026] FCA 1264, the court recognised that AI use for collation and synthesis, alongside sufficient verification, human supervision, and traceability to source material, could advance the overarching purpose. “If AI can properly be used by the plaintiffs to extract and organise information from a large body of documents, subject to appropriate verification, there is no principled reason why the defendants should not avail themselves of equivalent technology to expedite their own review, including the preliminary categorisation of contracts by type, subject to verification against the source documents,” Justice Cheeseman wrote for the court. “The overarching purpose is better served by both parties making appropriate use of available technology than by treating such tools as a basis either for criticising the party that uses them or excu