Speech delivered by the Hon Justice Emilios Kyrou AO at the Judging in the Digital Age Seminar, National Judicial Institute of Canada, Edmonton, 25 June 2026^*

Introduction

Australia is a federation, with the three branches of government operating at the Federal, State and Territory level. The High Court of Australia is the equivalent of the Supreme Court of Canada. It operates nationally, as do the Federal Court of Australia and the Federal Circuit and Family Court of Australia. Each of the six States and two Territories has a Supreme Court and lower courts. In addition, there are Federal, State and Territory administrative tribunals, the largest of which is my tribunal, the Administrative Review Tribunal (ART), which operates nationally.

There is no overall court or tribunal governing body, so each court and tribunal makes its own rules and issues its own practice directions. This means that there is no uniform national approach to the use of artificial intelligence (AI)¹, either by the relevant court or tribunal, or the parties to proceedings before them.

There is a non-statutory body called the Council of Chief Justices of Australia and New Zealand, which comprises the Chief Justice of the High Court of Australia, the Chief Justice of the Supreme Court of New Zealand, the Chief Justice of each of the Australian State and Territory Supreme Courts, the Chief Justice of the Federal Court of Australia and the Chief Justice of the Federal Circuit and Family Court of Australia.

In March 2026, the Council of Chief Justices published the fourth edition of its Guide to Judicial Conduct, which states the following in relation to to AI: 

Courts in the Australian jurisdictions have variously issued practice or guidance notes on the use of Generative AI by judges, parties, lawyers and expert witnesses, or are in the process of formulating guidance in this respect. Judges must follow the guidance issued in their respective jurisdictions.

Judges must not use Generative AI for judicial decision-making. The prohibition on AI use for judicial decision-making does not encompass supportive uses of AI. Judges should also encourage their personal staff to discuss their use of AI tools with the judge.²

Consistent with the Guide to Judicial Conduct, the use of AI in Australia can be considered conveniently in relation to the:

• use of AI by judges and tribunal members for decision-making;
• use of AI by judges and tribunal members, and their staff, for research and other non decision-making purposes;
• use of AI by courts and tribunals for administrative purposes; and
• use of AI by parties, lawyers and witnesses participating in court and tribunal proceedings. 

I will discuss each of these uses in turn, focusing on the concerns and challenges on the use of AI in the Australian court and tribunal system. I will then outline my own perspective on AI and why I do not knowingly use it for any personal or official purposes.

At the outset, it is worth noting that the use of AI by Australian courts and tribunals is not regulated by any special statutes, and we do not have any specific ‘AI Act’. Some existing statutes on areas such as privacy, discrimination, criminal law and copyright have been amended to address AI issues. Courts and tribunals have mostly addressed AI through their own practice directions and codes of conduct. Some of them have emphasised that lawyers must comply with their obligations in the practice rules of legal profession regulatory bodies.  

Use of AI by judges and tribunal members for decision-making

Insofar as Australian courts and tribunals have issued practice directions or guidelines on the use of AI, they have either remained silent on the use of AI for decision-making by judges or tribunal members, or have prohibited such use for decision-making.

For example, paragraphs 9.1 to 9.3 of the Code of Conduct that applies to non-judicial members of the ART provide as follows:

9.1  A Member must not use Generative AI to obtain guidance on the outcome of a proceeding, to produce any part of the Member’s reasons for a decision or to obtain any form of feedback or assistance on any part of the Member’s reasons for a decision which the Member has already prepared.

9.2  A  Member must not enter any Tribunal information, data or records, including case or party data, emails, reports, chat logs, code and system errors, into any Generative AI application.

9.3  Where a Member uses Generative AI as a general research tool without breaching the obligations in [9.1] and [9.2], the Member must check any research generated by Generative AI and verify its accuracy before relying on that research.³

These provisions are consistent with the Guide to Judicial Conduct of the Council of Chief Justices. They are also consistent with a recent report by the Victorian Law Reform Commission, which has recommended that guidelines for judicial officers should prohibit the use of AI for judicial decision-making, but not for supportive uses.⁴  

The prohibition on the use of AI for judicial and quasi-judicial decision-making reflects fundamental concerns and challenges regarding the use of AI by judicial and quasi-judicial decision-makers.

A key concern is to ensure that the decisions and reasons for decisions of courts and tribunals are entirely those of the independent and impartial judicial or quasi-judicial decision-maker. Each decision-maker is required to discharge their oath of office, namely, to ‘well and truly serve in the office, and do right to all manner of people according to law without fear or favour, affection or ill-will’. A decision-maker would act inconsistently with their oath of office if they shared their decision-making function with an AI application.

Another important concern is the extent to which information that is entered into AI applications will be kept confidential and secure. Having regard to the highly sensitive information that judges and tribunal members must frequently consider, there are serious privacy risks in using AI in decision-making. There is also the risk of material being included in AI generated content which breaches legal professional privilege, copyright or other intellectual property rights, or that the material is false or misleading.  Use of AI generated content could also result in inadvertent plagiarism of someone else’s work.

Although the Guide to Judicial Conduct of the Council of Chief Justices does not seek to prohibit what it describes as ‘supportive uses of AI’ by judges, it does not define what those uses encompass. There can be a fine line between using AI tools for research and improved readability, and AI inspired alterations which may ostensibly improve clarity but, on close analysis, alter legal meaning.

A key challenge on the use of AI by judges and tribunal members is its unreliability. It is well known that AI generated content (particularly from ‘public’ AI models) can be inaccurate or even completely fictitious. Judges and tribunal members are essential pillars of the rule of law, and their decisions profoundly affect the lives and livelihoods of citizens. Decisions which are based on flawed AI generated content represent not only sloppy work, but are highly detrimental. They can damage public trust and confidence in courts and tribunals, and undermine the rule of law.

In a paper delivered on 27 March 2026, the Hon Robert French AC, a former Chief Justice of the High Court of Australia, emphasised the following concerns with the use of AI:⁵ 

• Dispute resolution processes have an inescapably human dimension, which involves fundamental norms and values that reflect recognition of the essential dignity of the human person. Some applications of technology to those processes may lack those norms and values, and thereby diminish the centrality of human dignity and the power of people to give effect to it.
• Over-reliance on technology in the law may contribute to the progressive de-skilling of the human capacity to resolve disputes.
• The so-called ‘black box problem’ – the challenge in comprehending the precise mechanism by which an AI tool arrives at a particular outcome. From a transparency perspective, a question arises as to whether using an AI tool in resolving disputes is any different from resolving them by a throw of the dice.  

I am not aware of any cases where an Australian judge or tribunal member has been criticised for using AI in their decision-making.

Use of AI by judges and tribunal members for non decision-making 

As we have seen, the Guide to Judicial Conduct of the Council of Chief Justices and the ART Code of Conduct recognise that judges and tribunal members can use AI as a research tool for identifying potentially relevant material. However, it is essential that the relevant decision-maker personally checks that material and verifies its accuracy, reliability and relevance.

The reference to judges encouraging their personal staff to discuss the use of AI tools in the Guide to Judicial Conduct reflects the concern that judges need to know when their associates and other personal staff use AI in preparing material to assist the judge in their decision-making. It is obviously important that judges are aware of, and are able to control, the material that is provided to them by their staff.

Another concern about the use of AI by decision-makers and their staff is uncertainty about whether a particular research tool or software program incorporates undisclosed AI applications. In other words, can judges and tribunal members be entirely confident that they and their staff are not unintentionally using AI? For example, the Microsoft Copilot application can be used for proofreading of documents. However, unless care is taken, suggested corrections may extend beyond mere typographical errors and include changes to substantive content.

I do not think that AI has, to date, had a significant impact on the way that Australian judges and tribunal members personally perform their work during hearings or in preparing their decisions. There is a lack of uniformity among judges and tribunal members regarding the level of understanding of AI and concerns about it. To date, comprehensive AI training programs have not been developed for Australian judges and tribunal members to broaden their knowledge of how they may be able to harness its potential safely and responsibly.

Currently, AI is being used in limited ways during hearings, such as to facilitate the presentation of exhibits and other information. It is also being used to conduct research.

As discussed below, the use of AI by parties, lawyers and witnesses has necessitated that judges become more vigilant in checking submissions and other documents with which they are provided, to ensure that the documents do not contain false or misleading AI generated content.

Use of AI by courts and tribunals for administrative purposes

Judges and tribunal members are supported by public service staff who provide registry and corporate services. Those services include information technology, human resources, finance, records management and security. Most of these services are not visible to members of the public and do not directly affect them. Experience has shown that AI tools can produce operational efficiencies in the delivery of these services. It is therefore not surprising that it is generally recognised that the use of AI tools by courts and tribunals for administrative purposes is acceptable, provided that the tools are properly controlled and managed. In fact, the Australian Government is encouraging all its agencies to use AI for administrative functions.⁶ 

Examples of uses of AI for administrative purposes by the Federal Court of Australia include:

• Use of AI for the performance of some of the Court’s corporate services. At this stage, small scale limited trials of personal productivity uses are being undertaken, with a view to assessing possible future uses.
• Use of Microsoft Copilot by Court staff to transcribe internal meetings and produce meeting summaries.
• The Court has operated a digital court record since 2024. The Commonwealth Courts Portal, which enables the e-filing system, allows lawyers and parties to file court documents electronically in real time. The e-filing system lends itself to increased use of AI by parties and lawyers.  

It is fair to say that the Federal Court of Australia is proceeding cautiously in relation to the use of AI for administrative purposes. It is trialling, testing and assessing particular uses before deciding whether to roll them out systematically, due to the concerns and challenges about the use of AI to which I have already referred. 

The ART is at a similar stage in its exploration of AI. Examples of uses of AI for administrative purposes by the ART include:

• A coordinated trial for ART staff of premium Microsoft Copilot features, including specific AI agents (such as ‘Researcher’ and ‘Analyst’ assistants) to access ART information across all Microsoft products.
• Use of Microsoft Teams Facilitator to improve meeting productivity by automatically generating real-time notes, and tracking agenda and action items.
• Implementation of Microsoft Security Copilot to support cyber security functions, including enhanced threat detection and analysis, as part of broader efforts to strengthen the ART’s security posture in a controlled and secure way.
• Use of GitHub Copilot for Azure to support ART technical teams with development and configuration tasks within approved environments, with a view to improving efficiency, code quality and consistency.

As with the Federal Court of Australia, the ART is trialling, testing and assessing potential uses of AI before any broader rollout. This cautious approach reflects the concerns and challenges to which I have already referred. The ART is particularly keen to avoid inappropriate reliance on AI and to ensure that accountability and judgement remain with humans.

Use of AI by parties, lawyers and witnesses

Many parties, lawyers and witnesses have embraced AI in proceedings before courts and tribunals. Lawyers, in particular, have found that AI can improve efficiency and reduce costs in the performance of some common tasks. Those tasks include:

• research, particularly by using closed-source AI tools developed by legal publishers;
• improvement to the grammatical correctness and readability of documents;
• preparing presentation material, such as PowerPoint presentations;
• document review and management;
• discovery;
• collation, pagination and presentation of documents;
• preparation of indexes, summaries, chronologies, transcripts, and template emails, letters and notices;
• translation of documents; and
• e-filing of documents with courts and tribunals.

Self-represented litigants have relied upon AI as a means of reducing the knowledge gap between them and legally represented opponents. AI can assist them to draft submissions and other court documents that are more articulate and logical than the documents they would have been able to prepare on their own.

A key concern in relation to the use of AI by lawyers and parties – particularly self-represented parties – is that the AI generated content is sometimes not checked and verified. Unfortunately, some self-represented parties are unable to determine either accuracy or relevance. AI generated content that looks compelling to them is often irrelevant to their case or made up. Examples include citation of cases that do not exist or legal propositions that have no foundation in the caselaw.

A key challenge in relation to the use of AI by parties and lawyers is that it places the onus on the opposing party and their lawyers, and on judges and tribunal members, to determine what is real and what is hallucinatory. In the case of self-represented litigants, judges and tribunal members would ordinarily be alive to the need to check the accuracy of their material. Prior to the advent of AI, judges and tribunal members could usually rely upon the professionalism of lawyers to refrain from submitting material unless they had carefully checked it. Regrettably, that has not always been the experience since AI applications have become generally accessible. There have been several examples of Australian lawyers being referred for disciplinary investigation because they have used inaccurate AI generated material without checking it.⁷

AI can undoubtably impose a heavy burden on judges and tribunal members to check the accuracy and reliability of material they receive from parties and lawyers. Courts and tribunal members have some control over the use of AI by lawyers and other regulated professionals (such as migration agents) due to the disciplinary mechanisms that apply to them. Once a lawyer or another regulated professional has been disciplined for improperly using AI, they are unlikely to re-offend. However, there are fewer mechanisms available to courts and tribunals to prevent repeated improper use of AI by parties themselves.⁸ 

Ironically, courts and tribunals may have to turn to AI tools to assist them to assess the accuracy of material submitted by parties, and to detect instances of improper use of AI by them. In a speech titled ‘Litigants In Person and the Administration of Justice’ on 27 February 2026, the Chief Justice of the Federal Court of Australia stated that the Court was investigating the use of closed-source AI tools by the Court to collate and assess information provided by self-represented litigants, and produce coherent chronological statements of the claims submitted by them to the Court. Her Honour explained her rationale for this approach as follows: 

[T]he Court seeks not to deny the reality of the use of artificial intelligence by litigants in person, but to find ways to support its use that might be helpful to the articulation and resolution of proceedings. ...

... If litigants in person are going to use artificial intelligence, which in my opinion is now inevitable, it is better they do so with assistance and under conditions likely to produce documents that are as accurate and useful as possible.⁹

The use of AI by parties, lawyers and witnesses is not regulated by any special statutes. The general law dealing with matters such as breach of confidence and false and misleading statements applies. Courts and tribunals have issued their own practice directions to regulate the use of AI by parties, lawyers and witnesses. In relation to lawyers, longstanding professional standards – such as not to mislead a court or tribunal – apply to the use of AI. In addition, some professional regulatory bodies have issued specific guidelines dealing with the use of AI.¹⁰  

The Supreme Court of New South Wales was one of the first Australian courts to issue a practice note on the use of AI by parties and practitioners. Its practice note, which is titled ‘Use of Generative Artificial Intelligence (Gen AI)’, prohibits the inclusion of AI generated content in evidentiary documents and imposes verification obligations in relation to other documents that are submitted to the Court. It relevantly provides as follows: 

10.  Gen AI must not be used in generating the content of affidavits, witness statements, character references or other material that is intended to reflect the deponent or witness’ evidence and/or opinion, or other material tendered in evidence or used in cross examination. This paragraph does not prohibit the use of Gen AI for work that is merely preparatory to the drafting of the affidavit or other document setting out the witness’ evidence and/or opinion.
...
13.  An affidavit, witness statement or character reference must contain a disclosure that Gen AI was not used in generating: … its content (including by way of altering, embellishing, strengthening or diluting or rephrasing a witness’s evidence); …
…
16.  Where Gen AI has been used in the preparation of written submissions or summaries or skeletons of argument, the author must verify in the body of the submissions, summaries or skeleton, that all citations, legal and academic authority and case law and legislative references: 
(a) exist, 
(b) are accurate, and 
(c) are relevant to the proceedings,
and make similar verification in relation to references to evidence in written submissions or summaries or skeletons of argument to evidence (whether the evidence be contained in affidavits or transcript).¹¹ 

The approach of the Supreme Court of New South Wales can be contrasted to the less prescriptive approach of the Federal Court of Australia. The latter Court issued an AI Practice Note on 16 April 2026. It relevantly states as follows:

4.5  Some documents filed in a proceeding must contain the name of the person or lawyer responsible for preparing the document. If Generative AI tools have been used in the preparation of documents, the Court expects that the responsible person will have confirmed that:
(a) facts stated in pleadings are based on what the party reasonably considers can be proved and claims for relief are based on proper legal principles;
(b) legal authorities cited in submissions exist and support the proposition stated;
(c) evidence cited in submissions exists, is or will be in the materials before the Court and is reasonably likely to be admissible; 
(d) statements about what the evidence proves are findings reasonably open for the Court to make;
(e) chronologies are accurate; ...
...
4.10  The use of Generative AI must be disclosed where Generative AI tools were used:
(a) to summarise or analyse information upon which a witness relies to make a statement of fact or express an opinion;
(b) to create images, video recordings, sound recordings or other multimedia that are presented to the Court, which should be clearly identified as having been produced using Generative AI in the highly special circumstances where the creation of such materials using Generative AI has been considered necessary for some purpose relevant to the proceedings before the Court; or
(c) in any other manner that might reasonably be expected to affect the admissibility of that evidence, or what use is made of it by the Court.
4.11  Disclosure should occur in the body of the document at the start of the document. It should say as concisely as possible where in the document Generative AI has been used, and how it has been used. 
...
5.1  Where Generative AI is used in way that is inconsistent with this Practice Note or the Court’s orders or directions, all persons should expect that there could be consequences including adverse costs orders and issues as to compliance with legal and professional obligations.¹²

The use of AI by expert witnesses poses unique concerns and challenges. The documents that parties and lawyers provide to courts and tribunals deal largely with factual and legal issues. Judges and tribunal members are able to check them and form their own views about them. Expert witnesses, on the other hand, provide expert reports which can deal with highly specialised topics that may be outside the expertise of judges and tribunal members. Experts may also use specialised closed-source AI programs with which judges and tribunal members may be unfamiliar. For these reasons, judges and tribunal members are not as well equipped to detect errors or hallucinations in expert witness reports.

To address these concerns and challenges, many courts and tribunals have updated their practice directions relating to expert witnesses. The updated practice directions typically require expert witnesses either to seek leave to use AI in preparing their reports or to disclose whether they have done so and, where they have, to identify the AI applications they have used and verify the accuracy of the AI generated material.

For example, paragraphs 3.5B and 3.5C of the ART’s Expert Evidence Practice Direction provide as follows:

3.5B  An expert preparing a report must state in the report whether the report includes content generated by using Generative AI (AI content).
3.5C  Where an expert states that the report includes AI content, the expert must:
(a) clearly identify the AI content and the application(s) used to generate the AI content; and
(b) certify that the expert has personally checked all the AI content (including research and other material cited in support of the content) and is satisfied that it is all accurate and reliable.¹³ 

Personal perspective on AI 

I do not knowingly use AI for personal or official purposes. I do not understand it sufficiently and therefore feel uncomfortable in using it. I have the same approach to other products or services. For example, I do not invest in financial products that I do not understand. Of course, one of the risks in this area is that I may be using what I consider to be conventional research tools which in fact have embedded AI capability.

I support the position adopted by the Council of Chief Justices that judges should be prohibited from using Generative AI for judicial decision-making. The prohibition should also extend to tribunal members. There are two key reasons for my view.

The first reason is philosophical. Like former Chief Justice Robert French, I believe that humans need to operate within a values and ethical context, which AI lacks. Judges and tribunal members need to make many decisions which involve careful weighing up and balancing of competing considerations and value judgements which cannot be reduced to arithmetical formulas or algorithms.

Secondly, I share Robert French’s concerns about de-skilling. The more judges and tribunal members use AI, the greater the risk of them losing some of the decision-making skills that they have developed over many years. The problem of de-skilling is even more acute for young lawyers and newly appointed judges who become accustomed to using AI at the outset, before their own inherent decision-making skills fully mature.

At a personal level, I want my decisions to reflect my unique voice and style, which I have developed over many years in line with my growth as a jurist and my deeper appreciation of the foundational values of our legal system. I do not wish to share my voice with computer generated content which is based upon an instantaneous collation of random data that is likely to lack a core unifying value system.  

Thank you. 

^* I gratefully acknowledge the useful feedback I have received on a draft of this paper from Emeritus Professor Tania Sourdin, my Associate James Krotsis, Judicial Registrar Christine Cody of the Federal Court of Australia and the Principal Registrar of the Administrative Review Tribunal, Michael Hawkins AM.

¹ Unless otherwise indicated, all references to AI in this paper are to generative AI, which I will define as a system of artificial intelligence that is capable of generating ‘content’, such as text, images or sound, in response to prompts. It includes both open-source applications and closed-source applications.

² Council of Chief Justices of Australia and New Zealand, ‘Guide to Judicial Conduct’ (Australian Institute of Judicial Administration, 4th ed, March 2026), 13.

³ Administrative Review Tribunal, ‘Administrative Review Tribunal Code of Conduct for Non-Judicial Members’, 14 October 2024.

⁴ Victorian Law Reform Commission, ‘Artificial Intelligence in Victoria’s Courts and Tribunals Report’, October 2025, xx (recommendations 13 and 14). 

⁵ The Hon Robert French AC, ‘Stops on the Spectrum of Dispute Resolution’ (Keynote Address, International ADR Conference 2026, Sydney, 27 March 2026) [5]-[6], [18], [45], [53]-[54].

⁶ The Australian Government has issued a number of policy documents relating to the use of AI. For example, the 'AI Plan for the Australian Public Service 2025’ (at 12 November 2025), discusses improvement to government service delivery, policy outcomes, efficiency and productivity through substantial expansion of the safe and responsible use of AI in government. It summarises 15 initiatives across 3 pillars. 

⁷ Decisions of the Federal Circuit and Family Court of Australia include: Handa v Mallick [2024] FedCFamC2F 957, [7]-[10]; Dayal [2024] FedCFamC2F 1166; Valu v Minister for Immigration and Multicultural Affairs (No 2) [2025] FedCFamC2G 95, [32]-[38]; JNE24 v Minister for Immigration and Citizenship [2025] FedCFamC2G 1314, [26]-[35]. 

⁸ In Luck v Secretary, Services Australia [2025] FCAFC 26, [14], the Full Court of the Federal Court of Australia decided to redact the name and citation of a non-existent judgment that was relied upon by a self-represented litigant, to prevent ‘the false information [from being] propagated further by artificial intelligence systems having access to the [Court’s] reasons’. The decision of the New South Wales Court of Appeal in May v Costaras [2025] NSWCA 178, [1]-[17], [49], [95], illustrates how the use of AI by self-represented litigants can add to the costs and complexity of a proceeding and increase the burden on the other parties to the proceeding and the court. 

⁹ The Hon Debra Mortimer, Chief Justice, ‘Litigants In Person and the Administration of Justice’ (Paper, Commonwealth Law Conference, 27 February 2026) [56]-[57].

¹⁰ See, eg, Law Institute of Victoria, ‘Ethical and Responsible Use of Artificial Intelligence’ (at 13 August 2025). See also Ethics Committee, Victorian Bar, ‘Guidance on the Ethical Use of Generative AI’ (at 25 August 2025).

¹¹ Supreme Court of New South Wales, ‘Practice Note SC Gen 23: Use of Generative Artificial Intelligence (Gen AI)’. The Practice Note was originally issued on 21 November 2024, but was revised on 28 January 2025 following feedback from the legal profession (emphasis in original).

¹² Federal Court of Australia, ‘Use of Generative Artificial Intelligence (GPN-AI)’, 16 April 2026 (footnotes omitted). 

¹³ Administrative Review Tribunal, ‘Administrative Review Tribunal (Expert Evidence) Practice Direction 2026’, 27 January 2026.