Generative AI in litigation: from principles to responsible practice

Authored by Michael Kontoudis, Senior Legal Writer & Ruth Engelbrecht, Legal Writer (LexisNexis® Practical Guidance Dispute Resolution) & Vicki McNamara, Senior Research Associate (Centre for the Future of the Legal Profession, UNSW Law & Justice)

Generative artificial intelligence (Gen AI) has significant potential to improve the conduct of litigation.

To realise that potential, practitioners must ensure its use is grounded in verification, supervision, professional judgment, and expertise, while retaining responsibility for the material they rely on and the arguments they advance.

Gen AI is becoming embedded in legal practice. Practitioners can use it to accelerate their research, reduce the time and cost of routine and document-heavy tasks, and devote capacity to strategic analysis, case preparation, and client engagement.

Its rapid adoption has prompted Australian courts, tribunals and professional regulators to develop rules and guidance governing its responsible use.

The focus is now beginning to extend beyond managing the risks associated with Gen AI to identifying where it can be used effectively in litigation. In complex matters, assessing whether and how AI can assist may increasingly form part of effective case management.

Responsible AI adoption requires more than compliance with an evolving regulatory framework.

Effective use also depends on established practices of verification and supervision, the exercise of professional skill and judgment, and practitioners retaining responsibility for the AI-assisted work they produce.

The role for practitioners is to identify where AI can add value and ensure that it is used responsibly, subject to appropriate safeguards and in accordance with court requirements, professional obligations and firm protocols.

Three questions provide a useful framework for establishing matter-level controls and making task-specific decisions about Gen AI use:

1. Can we use it?
2. Can we rely on the output?
3. Who owns the result?

Listen: Ruth and Vicki unpack these three questions in the Legal Talk episode Gen AI in Litigation: Risk, Reward and the Role of Supervision.

Question 1: Can we use Generative AI?

There is no one-size-fits-all answer to whether Gen AI can be used in litigation. The boundaries of permissible use depend on the task and circumstances of the proceeding, including the jurisdiction, material involved, and any firm or client requirements.

What do Australian courts and tribunals require?

Australian courts and tribunals continue to develop a layered, jurisdiction-specific regulatory framework for the use of Gen AI. Relevant requirements may be found in court rules, practice directions, practice notes and guidelines.[1]

Do efficiency objectives support AI use?

In appropriate cases, the use of AI may support the efficient and proportionate conduct of litigation.

In complex, information-intensive litigation, where AI may be able to perform some tasks reliably and at lower cost, practitioners should consider whether conventional processes remain appropriate. Efficiency gains may be achieved by using AI to synthesise large volumes of information, prepare chronologies and issue-based summaries, and identify areas of agreement and disagreement to help narrow a dispute.[2]

Courts may require parties to explore the suitability of AI for particular tasks,[3] and its use may be subject to protocols agreed between the parties or ordered by the court.

What should matter-level controls cover?

Firms should implement internal matter-level controls with appropriate guardrails. They should be established early, where appropriate, reviewed as circumstances change, and proportionate to the risks of the proposed use.

Well-designed controls can support productive use by matching the safeguards to the particular task, information, and level of risk.

The following considerations should guide both matter-level controls and individual decisions about whether, and on what conditions, Gen AI is used:

• Applicable requirements: Identify which court requirements, firm policies and matter-level protocols and controls apply to the particular proceeding and proposed use, including whether the use is permitted and any conditions, restrictions, disclosure, or verification requirements.
• Nature of the proposed use: The permissibility of, and risks associated with, Gen AI use will vary by task. Administrative uses may present lower risks, while legal research and court-facing work may warrant greater safeguards. Some courts also impose specific requirements, restrictions or prohibitions for particular uses. For example, the Supreme Court of New South Wales prohibits the use of Gen AI to generate certain witness evidence and imposes additional requirements for its use in expert reports and written submissions.[4]
• Material provided to the tool: Practitioners should consider confidentiality, privacy, restrictions on use or disclosure and whether the tool’s data-handling arrangements are appropriate. Relevant restrictions may arise from statutory obligations, court orders and the Harman implied undertaking. Entering privileged material into open or publicly available Gen AI tools may give rise to waiver of legal professional privilege.[5]
• Client requirements: Clients may restrict Gen AI use, prescribe tools or data-storage arrangements, or require disclosure or consent.
• Tool selection: Use tools suited to the task, having regard to firm requirements and professional guidance. For legal research and analysis, tools that ground their outputs in authoritative, verifiable legal sources can provide a stronger basis for reliable use.
• Verification, supervision and sign-off: Confirm who is responsible for reviewing, supervising and approving AI-assisted work.
• Record-keeping and disclosure requirements: Consider how to track and record the use of Gen AI, and whether its use may need to be disclosed to a court or tribunal.

The practical question is therefore more nuanced: can this tool be used for this task, in this matter, with this information, and under what conditions?

Question 2: Can we rely on the output?

Even where the use of Gen AI is permitted, practitioners must assess the reliability of its output.

Particular caution is required when using general-purpose or open Gen AI tools for legal research or analysis, especially where the output is not grounded in authoritative, verifiable legal sources.

Gen AI can produce polished and plausible material quickly and at scale, making errors difficult to detect without proper verification.

Rules, matter-level protocols and approved tools are important guardrails, but reliable and productive use of Gen AI depends on established practices of verification, supervision and review, and the application of professional skill and judgment.

What does proper verification look like?

Outputs from any technological process adopted should be traceable to the underlying primary material and capable of being tested and corrected. The process should remain subject to appropriate human supervision.[6]

Practitioners must then verify those outputs by tracing factual propositions to underlying documents and legal propositions to authoritative legal sources.

They must check that those sources support the propositions advanced and are relevant and applicable to the matter.

The risks of relying on formal safeguards without effective verification are illustrated by Cork & Anor v Smith [2026] EWHC 1199 (Ch), in which a junior solicitor using a firm-piloted AI tool failed to check the underlying authorities despite the tool itself warning that verification was required.

The Court considered that this may have reflected a ‘misguided faith’ in the AI.[7]

What do reported cases reveal about Gen AI misuse?

The UNSW Centre for the Future of the Legal Profession’s GenAI, Fake Law & Fallout report examined 520 documented instances of Gen AI misuse in legal proceedings between January 2023 and September 2025 across 10 jurisdictions, including 87 cases in Australia. It highlights two related patterns:

• The risks extend beyond hallucinated authorities and include reliance on inaccurate material, flawed legal reasoning, irrelevant or inapplicable legal arguments, unfocused or excessive submissions, and procedural defects.
• Gen AI can amplify familiar weaknesses in legal practice, including inadequate supervision and training, over-delegation and insufficient review.

The research provides valuable insights into Gen AI use and misuse as examined in judicial decisions. Those decisions do not necessarily reflect the full range of Gen AI use or its effectiveness in legal practice more broadly.

Vicki discusses how the Centre's dataset has grown since the report, and what's driving the increase, in the latest Legal Talk episode.

Where AI-generated errors make their way into material relied on or put before a court, they often expose shortcomings in verification, supervision or professional judgment.

That is reflected in reported cases, where judicial scrutiny generally focuses on whether practitioners have met their professional obligations, rather than on the fact that Gen AI was used. More broadly, Australian courts have generally focused on regulating the responsible use of Gen AI rather than prohibiting its use altogether.[8]

Some courts have also expressly acknowledged its potential benefits.[9]

How do professional obligations apply to Gen AI use?

The fundamentals have not changed. Lawyers remain responsible for satisfying their professional obligations, including the obligation not to mislead the court, and exercising independent judgment about material they rely on or put before a court.

Where a legal authority is cited in a court document, practitioners must understand that authority and exercise professional judgment as to its scope, weight and application to the particular factual, legal and procedural context.

Where Gen AI is used, effective verification and supervision require an understanding of the tool and its limitations, together with independent checking of its output. Asking the same tool to confirm its answer is not independent verification.

The level of scrutiny should reflect both the immediate task and the intended end use. For example, an AI-assisted research memorandum or internal chronology may appear relatively low risk but warrant greater scrutiny if its contents later flow into court-facing material.

Question 3: Who owns the result?

Gen AI can add significant value to litigation work, but it cannot own the result. Professional duties and accountability remain with the practitioners who use, review, and rely on its output.

Failure to meet those professional obligations can have serious consequences, including personal costs orders, public judicial reprimands, regulator referrals and reputational harm.[10]

Some court regimes prescribe specific accountability measures, such as identifying a responsible person and requiring verification or disclosure where Gen AI is used.[11]

These AI-specific requirements operate alongside, and may add to, practitioners’ existing professional obligations. As the Chief Justice of the Federal Court of Australia has observed, the person who signs, files, relies on or presents a document to the court remains accountable for it.[12]

Who is accountable in a litigation team?

Accountability can become more complex in litigation teams, where several practitioners may contribute to the research, drafting, and review of a document.

Responsibility for AI-assisted work may arise at different stages, from use of the tool through to review, supervision, adoption and filing.

Supervising practitioners must understand how Gen AI is being used and provide appropriate oversight. They should not assume that responsibility rests only with the person who used the tool.

The identification of a responsible person, author or signatory does not confine responsibility to that person, nor does it negate the professional obligations of other practitioners involved in the relevant work.

Courts have referred multiple practitioners in the same litigation team to disciplinary bodies for their respective roles in a single AI-related error.[13]

From principles to responsible practice

While Gen AI tools and the rules governing their use continue to develop, the longstanding professional disciplines that underpin responsible legal practice remain fundamental.

Responsible legal practice also encompasses the efficient and proportionate conduct of litigation. In appropriate cases, where conventional processes are inefficient, responsible practice may require practitioners to consider whether AI could offer a more efficient approach.

Decisions about when Gen AI can be used, whether its output can be relied on, and who is responsible for the result must remain grounded in human verification, supervision, professional judgment, and accountability.

Beyond simply managing risk, these disciplines provide the foundation for confident, productive and responsible use of Gen AI in litigation.

LexisNexis Practical Guidance Dispute Resolution provides maintained guidance on Gen AI use in civil litigation, including:

• Guidance Note: Artificial intelligence and tech-assisted document review
• Toolkit: Comparative table — Obligations, duties and prohibitions on the use of Gen AI in Australian courts and tribunals
• Toolkit: Generative AI in civil litigation — Practice notes and guidelines

The Centre for the Future of the Legal Profession at UNSW Law & Justice conducts independent research on the future of legal practice, with a focus on areas including artificial intelligence, supervision and professional ethics.

[1] See Practical Guidance Dispute Resolution Toolkit: Comparative table — Obligations, duties and prohibitions on the use of Gen AI in Australian courts and tribunals.

[2] See R&B Investments Pty Ltd (Trustee) v Blue Sky Alternative Investments Limited (in liq) (Security for Costs and AI) [2026] FCA 1265 (R&B Investments) at [18]–[20], [23]–[28]; Rogers v McDonald’s Australia Ltd (AI-use) [2026] FCA 1264 (Rogers) at [10]–[16].

[3] See R&B Investments, order 6 and at [23]–[28]; Rogers, order 7 and at [15]–[16].

[4] See, eg, Uniform Civil Procedure Rules 2005 (NSW) r 31.4(3A)–(3C), 31.27(5)–(8) and 35.3B, and New South Wales Supreme Court Practice Note SC GEN 23 — Use of Generative Artificial Intelligence (Gen AI) at [10]–[25].

[5] Mertz & Mertz (No 3) (2025) 72 Fam LR 337; [2025] FedCFamC1A 222 (Mertz) at [15].

[6] See R&B Investments at [25]–[26]; Rogers at [13]–[14].

[7] Cork & Anor v Smith [2026] EWHC 1199 (Ch) at [38].

[8] See Practical Guidance Dispute Resolution Toolkit: Comparative table — Obligations, duties and prohibitions on the use of Gen AI in Australian courts and tribunals.

[9] See, eg, Federal Court of Australia, Use of Generative Artificial Intelligence Practice Note (GPN-AI) at [1.3]; Guidelines concerning the use of Generative Artificial Intelligence in Litigation in South Australian Courts at [5]–[6], [14]–[20].

[10] See, eg, Dey v Minister for Immigration and Citizenship (No 2) [2026] FedCFamC2G 1799 at [73]–[75], [77]–[82]; Mertz, order 1 and at [17]–[19].

[11] See Practical Guidance Dispute Resolution Toolkit: Comparative table — Obligations, duties and prohibitions on the use of Gen AI in Australian courts and tribunals.

[12] Federal Court of Australia, ‘AI Symposium’ (YouTube, 19 August 2026), https://www.youtube.com/watch?v=fRsbh_q6BE8, 00:09:48–00:10:00.

[13] See, eg, Pasuengos v Minister for Immigration and Citizenship (No 2) [2026] FedCFamC2G 96 at [33]–[38]; Mertz at [18]–[19].