AI Regulation Tracker / Courts and practice

Australian Federal Court Practice Note on Generative AI Use

On April 16, 2026, Chief Justice Debra Mortimer signed GPN-AI, a general practice note on the use of generative AI in Federal Court of Australia proceedings. According to the practice note published on the Court's website, the Court may require anyone who files documents with or appears before it to disclose whether and how generative AI was used, and the note sets out expectations around that disclosure. On evidence, the note asks that an affidavit or witness statement reflect the person's own recollection, and that generative AI used in preparing evidence be disclosed.

By Anthony Guerriero · The Leveraged Years Regulatory Research · Published April 16, 2026 · Last updated April 16, 2026 · Court practice note

The short version

The Federal Court of Australia has published GPN-AI, a general practice note governing generative AI in its proceedings, signed by Chief Justice Debra Mortimer on April 16, 2026. It is not new legislation. It is the Court setting its expectations under its own general powers to control practice and procedure. Two things stand out. First, the Court may require a person to disclose whether generative AI was used in a proceeding and how it was used. Second, when someone makes an affidavit or witness statement, the note expects that document to reflect their own recollection, knowledge, or experience, and asks that any generative AI used to prepare evidence be disclosed at the start of the document. The note is expressed to apply to everyone who files with or appears before the Court, including self-represented litigants, witnesses, and third parties served with a subpoena. It warns that hallucinated cases and citations are unacceptable, and that misuse can bring adverse costs orders. For US lawyers, expert witnesses, and companies caught up in Federal Court matters, the disclosure and evidence rules travel with the proceeding, not the passport.

Regulatory Briefing

Instrument: Use of Generative Artificial Intelligence Practice Note (GPN-AI), a general practice note of the Federal Court of Australia
Authority: Federal Court of Australia; issued by the Honourable Chief Justice D S Mortimer
Jurisdiction: Australia (federal courts); proceedings in the Federal Court of Australia
Status: In force. Signed and dated April 16, 2026. It reads together with the Central Practice Note (CPN-1).
Bindingness: Action. A practice note is not a statute, but it states the Court's expectations and disclosure requirements, and the Court can make orders and directions about generative AI use under its general powers in the Federal Court of Australia Act 1976 (Cth). Non-compliance can carry consequences, including adverse costs orders.
What Changed: Any person may be required by the Court to disclose whether and how generative AI was used in a proceeding. The note expects generative AI used to prepare evidence to be disclosed, and expects affidavits and witness statements to reflect the person's own recollection, knowledge, or experience.
Who This Affects: All persons who appear before or file documents with the Court, including represented and self-represented litigants, witnesses, experts, and third parties producing documents under subpoena. For US readers, anyone involved in Federal Court litigation, plus a comparative benchmark for US court AI orders.
Primary Source: Federal Court of Australia, GPN-AI, fedcourt.gov.au
Last Verified: July 23, 2026

The Federal Court has taken a position that a lot of courts are still circling. Rather than ban generative AI or wave it through, it has set expectations and attached a disclosure duty to them. The note opens by acknowledging that generative AI can help resolve disputes faster, cut legal costs, and widen access to justice. Then it spends most of its length on where the technology goes wrong and what the Court expects people to do about it.

The instrument names the tools directly. Paragraph 1.2 lists OpenAI's ChatGPT, Claude, Harvey, Google Gemini, and Microsoft Copilot as examples of generative AI. This is not an abstract policy about some future technology. It is aimed at the products lawyers and litigants are using right now.

What the Court actually expects

Three expectations sit at the center of the note. Any person who uses generative AI is expected to have a basic understanding of its capabilities, limitations, and risks. Any use must not adversely affect the administration of justice, which means users stay bound by their existing legal and professional responsibilities. And, in the operative line at paragraph 2.1(c), "If the Court requires it, a person must disclose to the Court if (and how) Generative AI has been used in a proceeding."

Read that carefully. Disclosure is not automatic across every filing. It is triggered when the Court requires it, and the note tells everyone to be ready. As paragraph 2.3 puts it, all persons are expected to be in a position to inform the Court what generative AI was used, how it was used, and for what purpose. The Court also reserves the right to require disclosure in other circumstances it considers appropriate, relying on its general powers under the Federal Court of Australia Act 1976 to make orders about practice and procedure.

The evidence line is the firm one

If there is a bright-line rule in GPN-AI, it is about evidence. Paragraph 4.8 states that any use of generative AI must be consistent with the requirement that when a person makes an affidavit or witness statement, they are representing that the document reflects their own recollection, knowledge, or experience. The note ties that to the criminal laws that prohibit falsifying or interfering with evidence. An expert report has to contain the expert's own opinion and reasoning, and experts keep their separate obligations under the expert evidence practice note.

Disclosure of generative AI is mandatory in specific evidentiary situations under paragraph 4.10: where the tool was used to summarize or analyze information a witness relies on to state a fact or give an opinion, where it created images, video, sound, or other multimedia presented to the Court, or where it was used in any way that might affect the admissibility of evidence or how the Court uses it. When disclosure is required, paragraph 4.11 says it belongs at the start of the document, stated as concisely as possible, saying where and how the tool was used.

Hallucinations are treated as a professional failure

The note is blunt about fabricated authority. It warns that generative AI can produce fictitious cases, citations, or quotes, incorrect statements of law, plain factual errors, and false confirmation that information is accurate when it is not. Presenting false or inaccurate information to the Court is described as unacceptable and inconsistent with the duty not to mislead. Where a person is responsible for preparing a document, the Court expects them to have confirmed that cited authorities exist and support the proposition, that cited evidence exists and is likely to be admissible, that chronologies are accurate, and that document lists conform to the Federal Court Rules. The message is that the person signing owns the output, not the model.

There is a confidentiality warning too. Paragraphs 4.13 to 4.15 caution that information entered into a generally accessible generative AI tool may become available to others, and that feeding privileged, suppressed, or otherwise confidential material into these tools can breach obligations, sometimes even with closed or ringfenced tools if outputs are later reused for a different purpose.

Why this reaches US practice

GPN-AI binds proceedings in an Australian federal court, not anyone in the United States by itself. It still lands on US desks in two concrete ways.

The first is direct exposure. If you are a US lawyer admitted or appearing in a Federal Court matter, a US company that is a party, or a US-based expert giving evidence in an Australian proceeding, these expectations apply to your conduct in that case. The disclosure duty and the evidence rules attach to the proceeding. A US expert who used a model to summarize source material behind an opinion, or a party that generated a demonstrative exhibit with AI, needs to be ready to disclose it in the form the note describes. Cross-border litigation teams should fold this into their engagement checklists now, because the obligation to be ready to explain your AI use is not something you can improvise at a hearing.

The second is comparative. US federal and state judges have been issuing their own standing orders and certification requirements on generative AI, and they vary widely. GPN-AI is a clean, well-reasoned model of one approach: no ban, a conditional disclosure duty, a hard rule protecting the integrity of sworn evidence, and personal responsibility for verifying anything a model produces. For US general counsel and litigators building an internal AI-in-litigation policy, it is a useful benchmark to read against your own courts' orders.

What to do now

If your work can touch an Australian Federal Court proceeding, keep a contemporaneous record of any generative AI used in preparing filings or evidence, including which tool, for what task, and by whom, so you can disclose it in the form the note requires if asked. Treat sworn evidence as off-limits for AI authorship: an affidavit or witness statement must be the person's own account. Verify every citation, quote, and factual claim in anything a model helped produce before it goes to a court, because the note puts that duty on the responsible person. And keep privileged and confidential material out of open AI tools unless you can be certain about how the tool stores and uses it.

Questions professionals are asking

Does GPN-AI ban lawyers from using generative AI?

No. The note accepts that generative AI can increase efficiency and access to justice. It sets expectations for responsible use instead of banning the technology. The core duties are to understand the tool's limits, to disclose use when the Court requires it, to keep AI out of the authorship of sworn evidence, and to verify anything a model produces.

When exactly must generative AI use be disclosed?

Disclosure is required whenever the Court requires it in a proceeding, and the note tells everyone to be ready to explain what was used, how, and why. Disclosure is also mandatory in specific evidentiary situations, such as where a tool was used to summarize or analyze information a witness relies on, or to create multimedia presented to the Court, or in any way that could affect admissibility. In those cases it must appear at the start of the document.

Can I use generative AI to help write an affidavit?

Not for the substance. Paragraph 4.8 requires that an affidavit or witness statement reflect the person's own recollection, knowledge, or experience, and it points to criminal laws against falsifying evidence. Where generative AI was used in preparing evidentiary material in the ways the note describes, that use must be disclosed.

Does this affect US lawyers or companies?

Not as domestic US law. It governs Federal Court of Australia proceedings. It reaches US readers in two ways: directly, when a US lawyer, expert, or party is involved in a Federal Court matter and must follow these expectations in that case, and comparatively, as a benchmark for US courts' own generative AI standing orders.

RELATED BRIEFINGS

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any practice note, statute, or requirement applies to your situation with qualified professionals in the relevant jurisdiction.