AI in dispute resolution in Australia: what courts and tribunals require

The short version

• Parties and lawyers may use AI in Australian courts, but the person who files stays responsible. Each court we checked has its own text.
• Evidence is where courts differ most. The NSW Supreme Court and the NSW Civil and Administrative Tribunal (NCAT) bar generative AI from writing affidavits and witness statements. South Australia’s guidelines say an AI first draft of a non-contentious witness statement may be appropriate.
• Since 20 October 2026, anyone who uses generative AI to prepare a Fair Work Commission document must say how they used it and confirm they checked it. Research for the Commission found that about 40% of surveyed applicants used AI to prepare or manage their case.
• Judges “must not use Generative AI for judicial decision-making” (Guide to Judicial Conduct, s 3.3).
• Australian arbitration and mediation bodies have no AI rules in force that we could find. Resolution Institute’s AI notetaking policy is a consultation draft, and we found no final version.
• From 10 December 2026, privacy policies must describe automated decisions, made or substantially helped by computer programs using personal information, that could significantly affect people (APP 1.7).^{1,3,13,18,20,21,22,26,28,36,37}

How is AI used in dispute resolution in Australia?

On the evidence we found, AI in dispute resolution in Australia is used mostly by parties preparing their documents. Institutions use it for support work rather than decisions. Artificial intelligence (AI) is a machine-based system that infers from its input how to produce outputs such as predictions, content, recommendations or decisions. Generative AI writes new text or images from a prompt (Supreme Court of Victoria, Practice Note SC Gen 25, 3.1).^7,22,25,32

Courts also see benefits. The Federal Court of Australia says generative AI could make litigation cheaper and widen access to justice (GPN-AI, 1.3). South Australia’s courts encourage lawyers and litigants to consider appropriate use (para 6).^1,13

Parties are the heaviest users we found. Research commissioned by the Fair Work Commission (FWC), which hears unfair dismissal and other workplace claims, surveyed 408 applicants and 211 respondents (President’s statement, 24 August 2026, [2]). That research found that about 40% of surveyed applicants used AI to prepare or manage their case ([11]).²²

It also found a spectrum from “AI-dependent” users, who tend to accept outputs at face value, to “AI-assisted” users, who check them. The more dependent an applicant, the more likely they are to arrive with inflated expectations and resistance to settlement ([14]). The Commission’s workload rose by over 70% in three years, and it infers this is “principally the result” of potential litigants’ growing use of AI tools ([1]).²²

The Administrative Review Tribunal (ART), which reviews federal government decisions, says it does not use and does not intend to use AI for its review decision-making (transparency statement, 18 June 2026). Its other uses of AI are subject to human oversight, and use of public AI is prohibited.²⁵

Among 23 financial services and credit licensees reviewed by the Australian Securities and Investments Commission (ASIC), AI was commonly used to triage complaints, and generative AI to summarise them was an emerging use (REP 798, p. 4 and Table 1, as at December 2023). Resolution Institute, a membership body for mediators in Australia and New Zealand, has drafted an AI notetaking policy and left “wider uses” of AI for later.^28,32

For how mediators, online tribunals, ombudsman schemes and institutions use AI elsewhere, see our guide to AI in dispute resolution. Our other regional guides cover Europe and South East Asia.

What do the Federal Court of Australia and state courts require when parties use AI?

Parties and lawyers may use AI in Australian courts, but they stay responsible for what they file. Each court we checked sets its own terms. The Federal Court of Australia issued its practice note on generative AI (GPN-AI) on 16 April 2026, after consultation.^1,2

GPN-AI covers anyone who appears before the Court or files documents with it, including people without a lawyer and witnesses (3.1). The Court may require disclosure of AI use whenever it considers it appropriate (2.1(c), 2.2), and disclosure is always required for some evidence (4.10). Where AI helped, the responsible person is expected to have confirmed that cited authorities “exist and support the proposition stated” (4.5). Misuse can bring “adverse costs orders” (5.1).¹

In New South Wales, the Supreme Court’s Practice Note SC Gen 23 has applied since 3 February 2025. It allows generative AI for chronologies, summaries and draft submissions (para 9B). Suppressed, subpoenaed or similarly protected material may go into a tool only if the person responsible is satisfied that it stays on a confidential platform, is used only for that proceeding and is kept out of model training (para 9A). The Court invited submissions on a review by 18 December 2025, but its AI page, last updated 23 March 2026, still lists the January 2025 version.^3,4,5

The Supreme Court of Victoria’s Practice Note SC Gen 25 commenced on 14 May 2026, replacing the Court’s 2024 guidelines for litigants. Users stay responsible “whether or not” they used AI (6.1) and must verify output with “meaningful human control” (7.1), keeping a human in the loop. Lawyers who rely on unverified output may face referral to their regulator and personal costs orders (6.7). The County Court of Victoria announced its own practice note on 9 June 2026; we read only the announcement.^7,9

The Supreme Court of Queensland’s Practice Direction 5 of 2025 requires the individual practitioner responsible for submissions to be named and to verify every reference (paras 5, 6, 9, 11). Self-represented litigants should read the courts’ guidelines for non-lawyers and try to check their references (paras 15, 16).¹⁰

South Australia’s Generative AI Guidelines apply from 1 January 2026 in six courts. The courts said in January 2026 that they “will shortly be given force” under court rules; the Uniform Civil Rules as compiled to 15 September 2026 do not yet refer to them. Using AI for a step in litigation is unlikely, by itself, to breach a lawyer’s or self-represented litigant’s obligations (para 8). No particular disclosure statement is needed, but users should be able to tell a judicial officer about their AI use if asked (para 43).^13,14,16

Can AI be used to write affidavits, witness statements or expert reports?

AI can help prepare evidence in the Australian courts we checked, but the evidence must reflect the witness’s own knowledge. The NSW Supreme Court and NCAT bar generative AI from generating its content, and both need leave before an expert uses it to draft or prepare a report. South Australia’s guidelines go furthest the other way and say an AI first draft of some witness statements may be appropriate.^{1,3,7,13,15,20}

The Federal Court requires an affidavit to reflect the witness’s “own recollection, knowledge and/or experience” (4.8). AI use must be disclosed at the start of the document where it summarised or analysed information a witness relies on, made multimedia or might affect admissibility (4.10, 4.11).¹

In the NSW Supreme Court, generative AI must not be used to generate the content of affidavits, witness statements and character references, though preparatory work is allowed (para 10). Those documents must state that it was not used (para 13). An expert given leave discloses which parts used AI (para 22). NCAT applies the same bar on AI-generated evidence and similar rules for expert reports (paras 12 to 14, 20 to 22), but its direction does not require a statement that AI was not used.^3,20

South Australia’s guidelines say AI can help by producing a first-person draft, with the witness’s consent, from an accurate transcript of the lawyer’s interview with the witness (para 53). Where the evidence is non-contentious and matches reliable documents, an AI first draft may be appropriate, though the lawyer should not prompt it to support a pleaded case and should review it with the witness (para 56). The guidelines call an expert opinion based on generative AI “likely to be inadmissible” (para 66).¹³

At VCAT, an expert may use AI to help prepare a report but not to generate an opinion (PNVCAT11, para 22). At the FWC, generative AI should not create a witness statement’s factual content, though it may transcribe or translate (paras 35, 36).^15,21

What do the Fair Work Commission and other Australian tribunals require on AI?

The Fair Work Commission’s Guidance Note is the most detailed tribunal text we read. Since 20 October 2026, anyone who uses generative AI to prepare a document for the Commission must say how they used it and confirm they checked the document (paras 8, 9). Forms in all individual rights matters now carry an AI section and a declaration that every user completes (President’s statement, [5], [8]).^21,22

The Guidance Note, published on 24 August 2026, does not cover AI used only for spelling, grammar or formatting (para 7). The checking must be done by a person; using the same or another AI tool does not count (para 31).²¹

Breaches can lead to documents being given less weight or disregarded, a costs order or the case being dismissed (para 15). Disclosing AI use does not change how the merits of a document are treated, provided the note is followed (para 26). Professional representatives who used generative AI must hyperlink cited cases; for others it is a recommendation (paras 11, 45).²¹

Other tribunals have texts of their own or rely on the courts’:^15,20,23

• ART: its President signed a Generative AI Practice Direction on 20 August 2026, covering when a person must verify, explain or disclose AI use. We read the Tribunal’s news item, not the direction.
• NCAT: Procedural Direction 7 has applied since 7 April 2025, based on SC Gen 23 with changes for the Tribunal (cover notes; para 1).
• Queensland Civil and Administrative Tribunal (QCAT) and the Queensland Industrial Relations Commission: the Queensland guidelines for non-lawyers and for judicial officers both apply there (non-lawyer guidelines, introduction; judicial guidelines [4]).
• Victorian Civil and Administrative Tribunal (VCAT): Practice Note PNVCAT11 has applied in all proceedings since 22 December 2025 (see the table above). The Victorian Law Reform Commission (VLRC) also recommended guidelines for members’ own use (rec 12).^{11,12,15,19,20,23}

For how online tribunals elsewhere handle AI, see our main guide.

Can Australian judges and tribunal members use AI?

Australian judges may use AI for supportive tasks but not to decide cases. The Guide to Judicial Conduct of the Council of Chief Justices of Australia and New Zealand (4th edition, March 2026) says judges “must not use Generative AI for judicial decision-making”. Supportive uses of AI are not caught (s 3.3).¹⁸

• New South Wales: no generative AI to write reasons, assess evidence, or edit or proofread draft judgments (paras 4, 5); judges may require litigants to disclose AI use (para 9).
• Supreme Court of Victoria: no generative AI for judicial decision-making (3.1); supportive uses such as summaries, chronologies, research and proofreading are allowed, with only Court-approved tools for case matters (3.4, 3.5).
• Queensland: no AI for decision-making or reasons ([25]).
• ART non-judicial members: no generative AI for guidance on an outcome, for any part of reasons or for feedback on drafted reasons, and no Tribunal information entered into it (Code of Conduct, 1.1, 9.1, 9.2).^6,8,12,24

The VLRC recommended the same line between deciding and supportive uses (recs 13, 14). For how judges in other countries use AI, see our guide to AI in law.¹⁹

What do Australian arbitration and mediation bodies say about AI?

We found no AI rules in force from Australian alternative dispute resolution (ADR) institutions, the bodies that run arbitration and mediation outside the courts. The Arbitration Rules 2021 of the Australian Centre for International Commercial Arbitration (ACICA) do not mention AI. Resolution Institute’s AI notetaking policy is a consultation draft.^26,28

Resolution Institute’s draft sits beside other mediation texts in our guidance table, and the same guide asks whether AI can replace a mediator. For institutional AI guidance and the enforcement of awards made with AI, see our arbitration guide.

How do Australian ombudsmen, complaint handlers and online services deal with AI?

The AI-specific complaint-handling guidance we found does not turn AI-written complaints away. The Ombudsman Western Australia tells agencies that “it does not matter if AI was used if you can understand the problem” (August 2026). Where a complaint is unclear, agencies should ask for the person’s own words. The Ombudsman’s own form now says “Tell us what happened in your own words”.³⁰

The same guidance notes that AI can help vulnerable people put a complaint into words. Signs of an AI-written complaint include very formal or legal language, a rigid structure, references to laws or policies that do not clearly relate, and incorrect facts.³⁰

ASIC’s Regulatory Guide 271 (2021), on how financial firms handle complaints internally, does not mention AI. It treats an expression of dissatisfaction about a firm’s products, services, staff or complaint handling as a complaint where “a response or resolution is explicitly or implicitly expected or legally required” (RG 271.27). A consumer need not use the word “complaint” (RG 271.30), so on our reading an AI-written complaint to a financial firm is still a complaint. In ASIC’s review published in October 2024, only 10 of the 23 licensees had policies referring to disclosing AI use to affected consumers (REP 798, p. 4).^31,32

In family law, National Legal Aid and the Legal Services Commission of South Australia run an online tool with Australian Government support. It uses AI to suggest how separating couples divide money and property. The VLRC calls it AI-assisted online dispute resolution (para 2.36).^19,33,34

The suggestion is not legally binding. The tool’s terms say it is not family dispute resolution, the formal process set out in the Family Law Act 1975 (Cth).^33,34

For the UK, see how complaint-handling rules apply to AI in our dispute avoidance guide. Our main guide covers how ombudsman schemes elsewhere use AI.

What does Australian law say about AI and automated decisions?

For AI in dispute resolution in Australia, the main new legal duty we found concerns privacy policies and automated decisions. Beyond that, the National AI Plan (2 December 2025) builds on existing, largely technology-neutral laws, with regulators addressing AI harms in their own areas and the government establishing an AI Safety Institute. The Plan points organisations to the Guidance for AI Adoption (21 October 2025), an updated and simpler version of the Voluntary AI Safety Standard.^40,41,42

The Privacy and Other Legislation Amendment Act 2024 (Cth) added APP 1.7 to 1.9 to the Australian Privacy Principles (APPs) in the Privacy Act. According to the Office of the Australian Information Commissioner (OAIC), the privacy regulator, they apply from 10 December 2026. From then, an entity’s privacy policy must say where it has arranged for a computer program to make, or do something substantially and directly related to making, certain decisions. These are decisions that use people’s personal information and could reasonably be expected to significantly affect their rights or interests.^35,36,37

The OAIC reads “computer program” broadly, covering pre-programmed rule-based processes as well as AI and machine learning (para 1.47). A decision can be in scope even where a person reviews the output (para 1.41). Among its examples of a program an entity has “arranged for”, the OAIC lists “a case management system that automatically escalates particular types of complaints”, which is in scope only where the decision is also significant.³⁷

Robodebt was a federal scheme that used income averaging to raise debts against welfare recipients; the Royal Commission into it reported on 7 July 2023. It found “no meaningful human intervention in the calculation and notification of debts” in the scheme’s Online Compliance Intervention phase (vol 2, pp. 476, 477). Recommendation 17.1 asks the Commonwealth to consider a consistent legal framework for automation in government services, with a clear path to review. Recommendation 17.2 asks it to consider a body to monitor and audit automated decision-making.^38,39

For the Commonwealth’s own AI use, non-corporate Commonwealth entities, with some exceptions, must meet the Digital Transformation Agency’s Policy for the responsible use of AI in government (version 2.0, from 15 December 2025). Its mandatory requirements include accountable officials and AI transparency statements, such as the ART’s.^17,25

The VLRC, reporting on Victoria’s courts and tribunals, called for a consistent and agreed national approach to AI regulation (rec 1). It also recommended that Victoria’s courts and VCAT tell people when a decision significantly affecting their rights was made or materially influenced by AI (rec 22). For the EU AI Act, see our guide to AI in dispute resolution in Europe.¹⁹

What happened when AI misuse reached Australian courts?

Australian courts have referred lawyers to their regulators and ordered some to pay costs personally.^43,44,46,48

AI output that looks real but is invented is known as hallucination. Some courts decline to reproduce invented citations so that they do not spread (Luck v Secretary, Services Australia [2025] FCAFC 26, [14]; JNE24, [26]). For how courts abroad have treated fake citations, see our guide to AI in law.^45,48

What should Australian parties, lawyers, mediators and complaint handlers do now?

Each step below follows from a text already in force, except the mediation step, which comes from a draft:

• Find out which court’s or tribunal’s text applies before using AI on evidence (see [the court and tribunal table](#courts)).
• Where AI helped prepare a document, check every citation yourself against a source known to be accurate; asking an AI tool to confirm it is not enough (SC Gen 23, paras 16, 17; SC Gen 25, 7.4(d), (e); FWC Guidance Note, para 31).
• Keep witness evidence in the witness’s own knowledge and words (GPN-AI, 4.8; SC Gen 25, 6.11). In the NSW Supreme Court and NCAT, do not use generative AI to generate its content, and get leave before an expert uses it (SC Gen 23, paras 10, 20; Procedural Direction 7, paras 12, 20).
• Keep suppressed, privileged and subpoenaed material out of AI tools unless the court’s conditions are met (GPN-AI, 4.13 to 4.15; SC Gen 23, para 9A; SC Gen 25, 6.8).
• Mediators: Resolution Institute’s consultation draft, not yet final, would require every participant’s written consent before AI notetaking, and deletion of AI notes and recordings within 24 hours after the mediation ends.
• In FWC matters, complete the AI section and declaration where the Commission’s form has one. If you are a professional representative who used generative AI, hyperlink every case you cite (President’s statement, [5], [8]; Guidance Note, paras 27, 45).
• Complaint handlers: the WA Ombudsman tells agencies to accept an AI-written complaint they can understand, to ask for the person’s own words when they cannot, and to consider giving complainants a simple prompt (August 2026).
• Entities bound by the Australian Privacy Principles: by 10 December 2026, make sure your privacy policy describes the kinds of personal information and decisions involved in automated decisions that could significantly affect people (APP 1.7, 1.8).^{1,3,7,20,21,22,28,30,36,37}

What is still open on AI in dispute resolution in Australia?

These questions were still open on 26 October 2026:

• What the review of SC Gen 23 will change, and when South Australia’s guidelines get force under court rules.
• What ACICA’s next Arbitration Rules say about AI, and whether Resolution Institute finalises its policy.
• Whether VCAT’s President issues guidelines for tribunal members (VLRC rec 12), and whether courts reach the national consistency the VLRC recommended through the Council of Chief Justices (rec 6).
• How APP 1.7 applies to the case systems of ADR bodies and complaint schemes.
• Whether the Commonwealth adopts the consistent legal framework for automation in government services that the Robodebt Royal Commission asked it to consider (rec 17.1).
• Whether the FWC’s inference holds that growing AI use by potential litigants principally explains the rise in its workload (President’s statement, [1]).
• What the ART’s direction and the County Court’s practice note say: we read only each body’s news item, not the texts themselves (how we research).^{4,9,14,16,19,22,23,27,28,36,38}