The Federal Court of Australia has today issued GPN-AI — Use of Generative Artificial Intelligence, a General Practice Note signed by Chief Justice Mortimer that sets out the Court's expectations for how Generative AI may be used in connection with proceedings.
The practice note is the outcome of an extended consultation process that began in 2025, when the Court established an AI Project Group and invited submissions from the legal profession, self-represented litigants, and the public. GPN-AI is the first formal Federal Court instrument dedicated to Generative AI, and it will sit alongside existing technology-focused practice notes such as GPN-TECH and GPN-EXPT.
Who it applies to
GPN-AI applies to all persons who appear before or file documents with the Court. That explicitly includes litigants (whether represented or not), witnesses, and third parties producing documents under subpoena. There is no carve-out for solicitors, barristers, or expert witnesses — the practice note speaks to everyone whose work ends up in front of the Court.
The Court's core expectations
The practice note sets out three baseline expectations: anyone using Generative AI should have a basic understanding of its capabilities, limitations and risks; use must not adversely affect the administration of justice and must be consistent with existing legal and professional responsibilities; and, where required, users must disclose to the Court how Generative AI was used.
The Court reserves the power to require disclosure in any circumstances it considers appropriate, using its general powers under the Federal Court of Australia Act 1976 (Cth). People appearing before the Court are expected to be in a position to explain what Generative AI was used, how, and for what purpose.
Where particular caution is required
GPN-AI identifies two areas of heightened risk.
The first is pleadings, written submissions, lists of documents, and other material lodged with the Court. The practice note directly addresses the hallucination problem: Generative AI "may give users fictitious cases, citations or quotes, or references to legal sources that do not exist". Anyone responsible for preparing such documents is expected to have confirmed that cited authorities exist and support the propositions made, that pleaded facts are grounded in what can reasonably be proved, that evidence cited in submissions exists and is likely to be admissible, and that chronologies and document lists are accurate. Presenting false or inaccurate information to the Court, even inadvertently via AI output, is described as unacceptable.
The second is affidavits, expert reports, and other evidentiary materials. Witnesses must be conscious that an affidavit represents their own recollection, knowledge or experience, and experts have an overriding duty to provide their own opinion and reasoning. Where Generative AI has been used to summarise or analyse material relied on by a witness, to create multimedia presented to the Court, or in any other manner that might affect admissibility, its use must be disclosed. Disclosure must appear at the start of the document, stating concisely where and how Generative AI was used.
The practice note also flags confidentiality as a separate risk area. Information that is the subject of suppression or non-publication orders, privileged, subject to the implied undertaking, or otherwise confidential should not be entered into a generally accessible Generative AI tool. The Court notes that even use of a "ringfenced" tool may breach obligations if outputs are later used for purposes outside the proceedings.
Consequences
The practice note is explicit that non-compliance can attract consequences, including adverse costs orders and issues with compliance with legal and professional obligations. That signal matters — it moves AI misuse from a reputational risk to a concrete procedural one that practitioners need to manage in their own workflows.
Our take
GPN-AI is a significant addition to the Australian AI governance landscape. It brings the Federal Court into alignment with the Supreme Court of New South Wales (which issued SC Gen 23 in late 2024) and the Supreme Court of Victoria's guidelines, while taking a somewhat more permissive position than either. Rather than prohibiting Generative AI outright in specific categories of work, the Federal Court has opted for an obligations-and-disclosure model that relies on existing legal and professional responsibilities to do most of the work.
That approach is consistent with how the Court has historically approached emerging technology, and it reflects a pragmatic recognition that Generative AI is already in widespread use across the profession. The emphasis on understanding capabilities, limitations and risks, combined with targeted disclosure obligations, puts responsibility squarely on the person in front of the Court — which is where legal responsibility has always sat.
For the broader Australian AI policy picture, GPN-AI is notable because it operates at the level of procedural rules rather than principles. Where documents like the DTA's AI PoC to Scale Guidance shape how government uses AI internally, GPN-AI shapes how AI may be used when interacting with a specific arm of the state. Both are part of a maturing Australian framework, and together they suggest that the operational phase of AI governance is now well underway.
You can read the full practice note on the Federal Court website.