Federal Court: Jahani v Qiu AI Costs Order for Generative AI

• The Federal Court in Jahani v Qiu [2026] FCA 1419 addressed cost implications after plaintiffs used AI to summarize over 7,500 documents in a corporate insolvency case.
• Defendants found errors in the AI-generated summaries, leading to a hearing adjournment from September 2026 to February 2027.
• The defendants sought a costs order based on the Federal Court Generative Artificial Intelligence Practice Note (GPN-AI), citing inaccurate AI material.
• The court denied the costs order, stating GPN-AI does not discourage responsible AI use with adequate verification and disclosure.
• Justice Cheeseman emphasized that AI for document collation and synthesis, with human supervision and traceability, advances the overarching purpose of litigation.

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.

The Federal Court of Australia recently addressed the cost implications arising from the use of artificial intelligence in a significant corporate insolvency case, Jahani v Qiu, in the matter of Ralan Property Services Pty Ltd (receivers and managers appointed) (in liq) (costs) [2026] FCA 1419. This complex proceeding involved corporate plaintiffs pursuing claims related to the alleged improper release and utilization of purchaser deposits by a group engaged in over 30 property development projects. The core dispute centered on claims of voidable transactions and accessorial liability, with the total unpaid released deposits reportedly exceeding $288 million at the time the group entered administration. The case also required extensive evidence concerning valuation, industry practices, and solvency.

In preparation for a scheduled hearing, the plaintiffs, on August 11, 2026, served proposed summaries derived from more than 7,500 documents, created with the aid of generative AI, or "super intelligence." Following an internal manual review and correction process by their solicitors, an updated version of these summaries, along with the underlying documents, was provided to the defendants on August 14, 2026. Upon examining this material, the defendants identified errors within the AI-generated summaries. While the plaintiffs acknowledged these inaccuracies, they disputed their significance to the overall case.

The plaintiffs subsequently filed an application under section 50 concerning the method of proving approximately 7,627 documents. An interlocutory hearing on September 22, 2026, was convened to address this application, where both parties concurred that the section 50 mechanism would be beneficial in avoiding the need to tender thousands of individual documents. However, the defendants raised concerns about procedural fairness, arguing that the summaries, served only weeks before the trial, left them insufficient time and access to source documents for proper verification amidst their ongoing trial preparations. Consequently, they sought to have the scheduled hearing vacated. The Federal Court, after reviewing the procedural history, decided to adjourn the partially heard section 50 application, revise the case timetable to allow parties to narrow their differences, and reschedule the hearing from its original September 28, 2026, start date to February 1, 2027.

Following the court's decision to adjourn the hearing, the defendants sought a costs order against the plaintiffs. Their application for costs was predicated on paragraphs 4.4 and 5.1 of the Federal Court’s Use of Generative Artificial Intelligence Practice Note (GPN-AI), which stipulate that the presentation of inaccurate AI-generated material could lead to adverse cost consequences. The defendants contended that the errors in the AI-assisted summaries, and the resulting delay, warranted such an order.

However, the Federal Court rejected the defendants' argument that the GPN-AI provided a distinct basis for imposing a costs order in this instance. The court clarified that the GPN-AI was not intended to deter the responsible application of AI tools for managing large volumes of documentary evidence, provided that such use is accompanied by adequate disclosure and thorough verification. The court found no impropriety in the plaintiffs' responsible use of AI for what it deemed a mechanical task involving an extensive number of documents.

Referencing the precedent set in Rogers v McDonald's Australia Ltd (AI-use) [2026] FCA 1264, the court underscored that the deployment of AI for tasks like collation and synthesis, when coupled with sufficient verification, human oversight, and the ability to trace information back to its original source material, could effectively advance the overarching purpose of litigation. Justice Cheeseman, in delivering the court's perspective, articulated that "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."

The ruling in Jahani v Qiu Federal Court AI costs provides crucial guidance for legal practitioners in Australia regarding the integration of artificial intelligence into litigation workflows. It signals the Federal Court's encouragement of responsible AI adoption, particularly for managing voluminous document summarization, rather than viewing it as a basis for criticism or an excuse for inaction. The court's stance emphasizes that while initial errors in AI-generated content may cause procedural delays, they do not automatically trigger adverse cost orders if the AI application is deemed responsible and aligns with the overarching objectives of the legal process.

This decision highlights the imperative for robust human verification, diligent supervision, and clear traceability to source materials when employing AI in legal contexts. The court's refusal to grant a costs order based solely on the GPN-AI provision, coupled with its emphasis on the "overarching purpose," suggests a pragmatic approach to technological advancement in the legal field. It reinforces that the benefits of efficiency gained through AI, when properly managed and verified, outweigh the risks of minor initial inaccuracies, thereby promoting the adoption of innovative tools to expedite complex litigation. The case serves as a clear directive for legal professionals to integrate comprehensive AI verification protocols into their practices, ensuring that technological advancements serve justice efficiently and fairly.

Lawyers and compliance officers should understand that the Australian Federal Court encourages the responsible use of AI for voluminous document summarization in litigation, provided there is adequate human verification, supervision, and traceability to source material. While initial errors in AI-generated summaries may cause delays, they do not automatically lead to adverse cost orders if the AI use is deemed responsible and serves the overarching purpose of litigation, suggesting a need to integrate robust AI verification protocols into legal workflows.

Source: Reporting based on legal news analysis