Artificial Intelligence in Litigation: Where Does Australia Stand?

Around the world, lawyers have been increasingly using AI to assist in litigation. As a result, many courts, have needed to address the permissible and impermissible uses for AI in litigation.

The response across Australia has been increasingly clear. AI is not prohibited, but responsibility cannot be delegated entirely. The lawyer or litigant who puts material before the court remains responsible for its accuracy. The position taken by different jurisdictions across Australia will be addressed and summarised in this article.

Position in Australia                  

The position courts have taken with respect to permissible use and guidelines in relation to AI differs slightly between jurisdictions and the particular court. However, the universal approach is that the output must be verified by the individual who is seeking to rely on the material/output. This is particularly relevant to litigation as AI has shown a tendency to refer to and generate case law which does not in fact exist. This issue has been identified in case examples which will be explored later in this article.

It is important to remember the difference between open and closed AI. Open AI systems make their models, code, or underlying technology more accessible for others to inspect, use, or modify, while closed AI systems keep these elements proprietary. The key difference is the level of access and transparency provided to users. For lawyers, the choice between open andclosed AI can have important implications for confidentiality and data security. Rather than relying on whether a system is simply “open” or “closed”, lawyers should consider how the AI provider handles client data, including whether inputs are retained or used for training, who can access them, where data is stored, and what safeguards apply to confidential and privileged information.

Federal Court of Australia

The Federal Court of Australia has now formalised its approach in the General Practice Note – Use of Generative Artificial Intelligence (GPN-AI), issued on 16 April 2026. The Federal Court recognises that generative AI can improve efficiency, reduce costs and enhance access to justice but emphasises that its use must remain consistent with existing legal and professional obligations.

The Federal Court Practice Note:

1. requires mandatory disclosure of the use of generative AI (in certain circumstances);

2. the ability for the Court to request further information in relation to the use of AI;

3. allows the use of AI in submissions, subject to verification;

4. allows the use of AI in affidavits and expert evidence; and

5. directs the guidance particularly to the use of open generative AI tools and notes that the use of closed tools can mitigate the risks of inadvertent disclosure.

NSW

NSW was an early mover in relation to the use of AI but is less encouraging of its use in comparison to the Federal Court. The Supreme Court of NSW’s Practice Note SC Gen 23 – Use of Generative Artificial Intelligence commenced in February 2025. In the Practice Note, the Court has adopted a relatively cautious approach, including significant restrictions concerning the use of generative AI in the preparation of evidence.

The Practice Note provides that:

- Confirms that generative AI does not include technology which merely corrects spelling or grammar, provides transcription or translation, assists with formatting and otherwise does not generate substantive content.

- Confirms that the Practice Note applies to both close and open-source generative AI.

- Information which is subject to suppression/non-publication orders, the Harman undertaking, subpoena material, or statutory publication restrictions must not be entered into generative AI tools.

- Generative AI must not be used to generate, alter, embellish, strengthen, dilute or rephrase affidavits, witness statements, character references or other evidence.

- Affidavits, witness statements and character references must state that generative AI was not used to generate their content, subject to limited exceptions.

- The use of generative AI to prepare an annexure or exhibit for evidence requires leave of the Court.

- The use of generative AI in submissions is not explicitly permitted but is otherwise not prohibited subject to verification in the body of the submissions that all citations and authorities exist, are accurate and are relevant to proceedings.

- Lawyers remain responsible for the accuracy and integrity of material prepared with AI.

Victoria

Victoria has since taken a somewhat more permissive approach. The Supreme Court of Victoria’s Practice Note SC Gen 25, commencing on 14 May 2026, permits court users to use AI in preparing court documents provided they comply with the Practice Note. Importantly, AI-generated content must be subject to meaningful human control and verification. Court users must verify authorities against reliable sources and may be required to identify material produced using AI and explain how it was verified.

The Practice Note in Victoria allows for the use of generative AI in the preparation of affidavit materials, witness statements and other documents but clarifies that caution must be used and the information must be verified with ‘meaningful’ human control.

The Practice Note also provides that confidential or sensitive information should not be entered into public AI (or, open AI) tools. For closed AI systems, lawyers must check the contractual terms and privacy settings to ensure information remains secure.

While the Practice Note applies to all ‘Court users’, additional obligations are imposed on lawyers including that the use of AI tools cannot be used as a substitute for their obligations to the Court and to provide independent and verified legal advice to clients.

Queensland

Queensland has taken a more targeted approach. The Supreme Court of Queensland’s Practice Direction 5 of 2025 addresses the accuracy of references in submissions and the risks associated with the use of AI. Similar directions have been introduced for the District Court and Planning and Environment Court, while Queensland's guidance for non-lawyers expressly warns that AI cannot reliably understand the specialised legal or factual context of a particular case. Interestingly, Queensland has amended its practice direction concerning expert evidence to require experts in relevant criminal proceedings to disclose the use of generative AI in formulating or expressing opinions.

South Australia

South Australia has issued Guidelines concerning the use of Generative Artificial Intelligence in litigation in South Australian Courts. Western Australia has the Supreme Court of Western Australia’s Guidelines for the Use of Generative Artificial Intelligence.

Australian Capital Territory, Northern Territory and Tasmania

The position is less developed in the remaining jurisdictions. The Australian Capital Territory and Northern Territory have not yet adopted an equivalent dedicated superior-court practice note comparable with NSW, Victoria or Queensland. Tasmania similarly does not have a dedicated AI protocol, neither Practice Note or Guideline in relation to the permissible or impermissible use of AI.  

The result is an increasingly important patchwork of approaches across Australian courts.

Case Examples

One of the earliest Australian cases to confront the issue of the use of generative AI was DPP v Khan [2024] ACTSC 19. During sentencing proceedings, Mossop J considered a character reference that appeared to have been generated or substantially rewritten using a large language model. His Honour considered that use inappropriate because it became difficult for the Court to determine what weight could be given to the facts and opinions expressed in the reference.

The case illustrates an important distinction: the problem is not simply that AI may produce a false case citation, but that AI can also alter the evidentiary character of a document. A character reference, witness statement or affidavit is intended to communicate a person's own knowledge, observations and opinions. Replacing that person's words with machine-generated prose can make it difficult for a court to assess what the witness actually knows or believes.

The NSW Court of Appeal confronted the issue more directly in May v Costaras [2025] NSWCA 178. The respondent, who was self-represented, used AI to prepare written and oral submissions. The submissions contained numerous irrelevant authorities and at least one ‘hallucinated’ case. Bell CJ warned of the need for judicial vigilance and observed that unverified AI use can increase the cost and complexity of proceedings without appreciable benefit.

The Federal Court has subsequently seen a similar problem in JML Rose Pty Ltd v Jorgensen (No 3) [2025] FCA 976, where a self-represented litigant used generative AI to prepare written and oral submissions containing hallucinated case references, legislation and Court rules. The case is a useful reminder that the problem is not confined to lawyers and that while AI can make litigation more accessible, but it can also make poorly founded arguments considerably more elaborate.

One of the best-known international example is the case of Mata v Avianca Inc which took place in the United States District Court for the Southern District of New York. In this case, lawyers filed submissions containing fictitious authorities generated by ChatGPT and were sanctioned.

Key Takeaways

Ultimately, AI can assist with and be a useful tool in litigation, but it cannot assume responsibility from any individual or legal advisor.

For practitioners, that means treating AI output as an unverified starting point rather than an authority. Every case citation, quotation, statutory reference and factual proposition needs to be checked and verified. Confidential and privileged material also requires particular care before being entered into an AI system.

The more fundamental question is whether AI is being used for a task that requires the lawyer's own professional judgment. There is an obvious difference between asking AI to suggest headings for a chronology and asking it to draft a witness statement, formulate an expert opinion or determine the legal basis of a claim.

Australian courts largely seem to be moving towards a middle ground, though there is presently no unified approach.The question is no longer whether lawyers will use AI but whether they will use it in a way that preserves the fundamental obligations of accuracy, confidentiality, competence and, above all, the duty to the court.

‍

Disclaimer

The above is general commentary only and is not legal advice. It is also not an exhaustive statement of the law, nor is it specific to your circumstances. If you are considering using artificial intelligence in Court proceedings or other legal contexts, or would like to discuss its use in your particular circumstances, you should seek independent and tailored legal advice.

‍