The Federal Court’s GPN-AI: Practical Lessons for Lawyers
The Federal Court’s Use of Generative Artificial Intelligence Practice Note (GPN-AI), issued by the Chief Justice on 16 April 2026, does not ban lawyers from using Gen AI. Instead, it says lawyers may use it carefully, responsibly, and consistently with their existing duties to the Court, their client, and the administration of justice.
The central message is simple: AI can help with efficiency, but the lawyer remains fully responsible for accuracy, confidentiality, and integrity.
The Why
The Court recognises that Gen AI can reduce cost, improve efficiency, and support access to justice. But it also sees real risks: false authorities, wrong legal analysis, factual mistakes, misuse of confidential material, and distortion of evidence.
In simple terms, AI may be fast, but professional responsibility remains with the lawyer. A useful way to think about it is this: AI is like a very fast junior assistant that can draft quickly, but can also confidently invent things.
Top Practical and Actionable Points for Lawyers
1. You can use AI, but you must understand its limits
The Court expects anyone using Gen AI in a proceeding to have a basic understanding of what it can do, what it cannot do, and what risks it creates. The Court can also require a person to explain whether AI was used and how it was used.
Simple example:
A solicitor uses ChatGPT to draft a first outline of submissions. That is not automatically improper. But if the solicitor does not check the authorities, facts, and reasoning, it becomes a professional problem.
Practical action:
- What tool was used?
- What was it used for?
- Was the output independently checked by a lawyer?
2. Never file AI-generated content without checking every authority, fact, and proposition
This is probably the most important operational point. The Practice Note expressly warns that AI may generate fake cases, fake citations, misleading legal propositions, factual errors, and even false reassurance that the answer is correct. The Court makes clear that presenting false or inaccurate material is unacceptable.
Simple example:
You ask AI: “Find me three Federal Court cases supporting this proposition.” It gives you three neat-looking citations. Two are real. One is invented. If you file that without checking, the problem is not that the AI made a mistake. The problem is that the Court was misled.
Practical action:
- Check that every case exists
- Check that every quote is real
- Check that every pin cite is correct
- Check that every legal proposition is supported
- Check that every factual statement matches the evidence brief
3. The lawyer signing or filing remains accountable
The Court makes clear that where documents are filed under a lawyer’s or party’s name, the responsible person must have confirmed key matters. That includes whether the facts are properly provable, the claims are legally sound, the authorities support the proposition, the evidence exists and is likely admissible, the inferences are open, the chronology is accurate, and discovery lists comply with the Rules.
Simple example:
You use Copilot to help build a chronology. It rearranges dates and accidentally places an email before the contract was signed. If you rely on it without checking, the chronology is wrong and responsibility still sits with you.
Practical action:
- Check pleadable facts against source documents
- Check causes of action against current law
- Check the chronology against original evidence
- Open and read the cases yourself
- Compare quotations against the source judgments
- Check discovery lists against the Rules
4. Be especially careful with affidavits, witness statements, and expert reports
The Court draws a sharper line around evidence. A witness statement or affidavit is supposed to reflect the witness’s own recollection, knowledge, and experience. An expert report must reflect the expert’s own opinion and reasoning process. This makes AI use in evidence preparation much more sensitive than AI use for administrative drafting.
Simple example:
Improper use: “Please rewrite this witness’s rough notes into a stronger, more persuasive affidavit.”
Why this is risky: AI may subtly improve, reshape, or add emphasis in a way that no longer reflects the witness’s actual memory.
Better use: “Please correct grammar in this draft, without changing substance,” followed by careful human review with the witness.
Practical action:
Use this rule of thumb: AI may assist with form, but must not manufacture substance.
5. In some evidence situations, disclosure is mandatory
The Practice Note says AI use must be disclosed where it was used:
- to summarise or analyse information relied on by a witness,
- to create images, audio, video, or other multimedia tendered to the Court, or
- in any other way that might affect admissibility or how the Court uses the evidence.
The disclosure should appear at the start of the document and should briefly explain where AI was used and how it was used.
Simple example:
A lawyer uses AI to summarise 500 pages of medical records for an expert, and the expert relies on that summary. That use should be disclosed.
Likewise, if a party tenders an AI-generated reconstruction image, it should be clearly identified as AI-generated.
Practical action:
Use a short disclosure statement such as:
“Generative AI was used only to summarise source materials identified at paragraphs X–Y. The deponent/expert reviewed the underlying materials and confirms the substance of this document reflects their own evidence/opinion.”
6. Do not paste confidential, privileged, or restricted material into public AI tools
This is another major operational rule. The Court warns that material entered into generally accessible AI tools may become available to others, and users may not know where it is stored, how it is used, or who can access it. This includes material that is confidential, suppressed, privileged, subject to an implied undertaking, or otherwise private.
Simple example:
You upload subpoenaed bank records into a public chatbot to “summarise the key transactions.” That could breach confidentiality obligations or the implied undertaking attaching to litigation material.
Practical action:
- Green: public material, procedural templates, neutral drafting help
- Amber: client material only in a vetted closed environment
- Red: never place privileged, suppressed, subpoenaed, confidential, or implied-undertaking material into ordinary public AI tools
7. Even “closed” AI systems are not automatically safe
The Practice Note notes that risks may be lower in controlled or ringfenced tools, but it still expects lawyers to preserve confidentiality rigorously. Even if the input environment is closed, later use of the output may still create problems if it breaches the purpose for which the information was originally obtained.
Simple example:
Your firm uses an enterprise AI product. That is better than a public chatbot. But if you feed it documents produced on discovery and then repurpose the output beyond the permitted use in the proceeding, there may still be a breach.
Practical action:
- Is the information privileged or restricted?
- What are the contractual and technical controls?
- Are we using the output only for the permitted litigation purpose?
8. There can be real consequences for getting this wrong
The Practice Note says misuse of AI may lead to consequences including adverse costs orders and issues concerning compliance with legal and professional obligations. The risk is therefore not theoretical. It can affect the case, the client, and the lawyer personally.
Simple example:
If a firm files submissions containing invented cases or tainted evidence preparation, the Court may do more than criticise the conduct. It may impose cost consequences and raise professional responsibility issues.
Recommendation
For a practical law-firm approach, the entire Practice Note can be reduced to one operating rule:
Use AI for speed, not for trust.
Let AI help with drafting, summarising, structure, and brainstorming. But do not let it become the final source of truth for law, facts, evidence, or confidentiality decisions. That judgment function remains human.
A sensible internal policy would say:
- Allowed uses: first drafts, structure, proofreading, and neutral summaries of non-restricted material.
- Restricted uses: anything touching pleadings, submissions, chronology, discovery, or evidence requires lawyer verification.
- Mandatory disclosure: where AI affects evidence in the ways identified by the Practice Note.
- Prohibited uses: putting privileged, confidential, suppressed, subpoenaed, or implied-undertaking material into public tools.
- Audit trail: record the tool used, purpose, inputs, reviewer, and verification steps.
That framework aligns closely with the Court’s expectations.
Very Simple Good / Bad Examples
Good:
“Draft a neutral case summary of this published judgment.”
Then the lawyer reads the judgment and checks the summary.
Bad:
“Write my submissions and give me authorities,” followed by filing without checking the cases.
Good:
“Suggest a cleaner structure for these submissions.”
Then the solicitor checks every proposition and citation manually.
Bad:
Pasting confidential discovered documents into a public chatbot to save time.
Good:
Using AI to help organise source material for an expert, then disclosing that use where required and ensuring the opinion remains the expert’s own.
Bad:
Using AI to rewrite a witness statement so it sounds clearer, stronger, and more persuasive, where that changes the witness’s actual recollection.