I INTRODUCTION

On the 16th of April 2026, the Federal Court of Australia released a new practice note – ‘Use of Generative Artificial Intelligence Practice Note (GPN-AI)’, in an earnest attempt to ensuring that professionals use Generative AI ‘appropriately, responsibly and with due care’.[1] It was just over a year ago when the NSW Supreme Court attempted to respond to this same technology in similar vein, through its Practice Note SC Gen 23, issued in November 2024 and commencing in February of the following year. It is worth considering that SC Gen 23 was one of the earliest comprehensive Australian court responses to the revolutionary yet polarising tool that is Generative AI, though it was initially met with mixed responses. The release of GPN-AI therefore invites the broader question as to SC Gen 23’s effectiveness – has it provided viable, workable guidance, or was it just a knee jerk reaction to a developing technology?

II WHAT DID IT PROPOSE?

SC Gen 23 is best defined as a risk-management instrument. It does not seek to arbitrarily prohibit, but rather, to regulate, permitting the use of Gen AI for largely administrative tasks, such as chronologies, indexes, witness lists, preparation of briefs, document and transcript reviews, and drafting of written submissions or argument summaries. [2] It considers generic language models, ‘bespoke’ legal programs and chatbots within these programs to be Gen AI,[3] but not established search engines or dedicated legal research software.[4] However, while its goal is not to prohibit, it nonetheless places safeguards against Gen AI for sensitive material, restricting its use in drafting affidavits, witness statements, or other evidentiary material, unless an applicant identifies the specific Gen AI tool used and is able to explain its beneficial impact, in which case special leave may be granted.[5] While Gen AI may assist in drafting written submissions, legal authorities cited must be correctly referenced and verified.[6] Similarly, the report also restricts AI-generated expert reports, requiring court approval before usage, and after its creation, requiring users to disclose the extent to which AI was used in preparation of the report.[7]

III HOW WAS IT RECEIVED AT THE TIME?

The practice note was seemingly met with overwhelming support from legal professionals. Sharon Kermath from Svenson Barristers, noted that GenAI had the potential to ‘improve the efficiency and quality of legal services, enhancing outcomes for clients and allowing lawyers to focus more on client relationships,’[8] while Law Society of NSW president, Brett McGrath, lauded its ability to ‘protect litigants, the broader community and the justice system itself’.[9]

It was supported by scholars, too. Mimi Zou, Head of the School of Private & Commercial Law at UNSW Law & Justice was welcoming to the practice note, expressing that it provides clarity and guidance what is still an emerging technological development.[10] Understandably, however, it was met with contention by key stakeholders in the AI industry, such as Samuel Junghenn, CEO of AI Legal Assistant, who noted that the practice note ‘lump[ed] together general-purpose AI chatbots like ChatGPT with purpose-built enterprise-level applications designed specifically for legal practice’.[11] Other critics also suggested that it risked stifling innovation in the legal sphere.[12]

IV HOW HAS IT IMPACTED LEGAL PRACTICE TODAY?

The biggest impact has been cultural, rather than punitive. SC Gen 23 forced legal professionals to confront the reality of using generative AI. While it has been incredibly convenient in saving time for certain tasks, it also has several drawbacks. The most recurring themes, as noted by the Centre for the Future of the Legal Profession, were fake and incorrect case citations and summaries (or hallucinations), flawed legal reasoning, incorrect documentation and prolix documentation and flooding.[13]  particularly its creation of hallucinations with alarmingly high confidence. The first domestic case in which a solicitor was caught making such a mistake was Dayal [2024] FedCFamC2F 1166,[14] and in recent times, JNE v Minister for Immigration and Citizenship [2025] FedCFamC2G 1314.[15] By ‘drawing a line’ around evidence, submissions and expert work, the Court made verification a professional habit, rather than an optional precaution. It also sent a clear message that AI is not a substitute for legal research, acting as a deterrent for professionals who may lack technological literacy, to perhaps not use it at all when drafting vital documents.

V HAS IT WORKED?

The most realistic answer is that SC Gen 23 has been of relative effect. It has clearly worked in signalling the Supreme Court’s expectations, normalising verification as non-negotiable and providing practitioners with a framework to understand when AI is permissible, and when it is not, and creating the understanding that there are repercussions for those who misuse it. The note is workable for larger firms with greater budgets that have integrated internal review systems. It is, however, more difficult to traverse for smaller practices and sole practitioners, as the note assumes an equal level of technological literacy that not all practitioners have, as well as self-represented litigants. This concern was noted by the Law Society Journal, which noted the risk of widening the digital divide between well-resourced firms against everyone else, furthering the structural gap.[16]

 It also makes the mistake of being tailored particularly towards qualified solicitors, despite there being numerous cases of self-represented litigants using Gen AI. This was demonstrated in May v Costaras [2025] NSWCA 178, whereby a self-represented respondent was found to have used Generative AI in her oral submissions, [17]  and Bottrill v Graham & Anor (No 2) [2025] NSWDC 221, where the second defendant used Gen AI for written submissions, and ‘valuable court time’ was ‘wasted’,[18] yet with no repercussions. This highlights a clear tension that the risks that motivated SC Gen 23 are not confined to solicitors, yet the regulatory burden falls most heavily on them. This asymmetry in enforcement means that practitioners face professional and reputational consequences for misuse of Gen AI, whereas self-represented litigants are more likely to be treated with fairness in the interests of fairness and access to justice. This also creates issues with the operation of key legal instruments such as the Civil Procedure Act 2005 (NSW) and its overriding purpose. Although it assists in the ‘just’ and ‘cheap’ resolution of issues, it cannot ensure that proceedings are ‘quick’, with scrutiny being required to ensure that there has been no misuse of Gen AI. Evidently, that SC Gen 23 operates more as a professional conduct framework and less of a universal safeguard.

However, while the practice note sets clear expectations, they may not be futureproof. The general philosophy of SC Gen 23 (which the Federal Court’s GPN-AI also shares) was that Gen AI can be useful, but users need to understand its capabilities and risks. SC Gen 23 attempts to manage these risks through ex ante control, imposing categorial restrictions. Contrastingly, GPN-AI reflects a shift towards ex post accountability, relying on professional obligations and the Court’s supervisory powers to address misuse if it were to arise, acting as a reminder for professionals of their responsibilities, including the accuracy of materials presented to the court.

VI HOW DOES GPN-AI IMPACT THIS?

Despite bearing striking similarities in the way it recommends Gen AI to be applied, the Federal Court’s GPN-AI does not simply replicate the approach taken in SC Gen 23. Rather, it represents a subtle, but important shift in regulatory philosophy. GPN-AI confirms the concern held in SC Gen 23 – that generative AI may produce inaccurate or entirely fabricated legal material, and that these risks are incompatible with the administration of justice.

SC Gen 23 takes a prescriptive, prohibition-based model that expressly restricts the use of Gen AI in certain scenarios. However, GPN-AI takes a more principles-based approach, grounded in existing professional obligations, focusing on outcomes and responsibilities. It emphasises that Gen AI may enhance efficiency, reduce costs and improve access to justice, but should be used cautiously in alignment with existing professional obligations, particularly the duty to not mislead the court and to verify legal authorities. It also prioritises transparency, allowing the court to demand disclosure of how AI tools were used and to the extent to which they were used. Although it may seem that GPN-AI is a ‘toothless tiger’ by not explicitly prohibiting the use of Gen AI, it is also understandable for a profession where solicitors are generally aware of the professional rules and obligations set out before them, even if they choose not to follow them.

This shift ultimately highlights an important limitation of SC Gen 23. While its prescriptive rules provide clarity, they may lack flexibility in a rapidly evolving technological landscape. Conversely, GPN-AI reflects an adaptive approach, recognising that AI is already embedded in legal practice, and is better managed through responsibility rather than outright restriction. Nonetheless, it is important to note that SC Gen 23’s prohibitive nature still bears value. Recently, scholars have undertaken the idea that possible solutions are ‘education, certification, sanctions and/or prohibition’, which do not exist in isolation, but work in tandem to combat issues caused by GenAI.[19]

VII CONCLUSION

Though SC Gen 23 may not have ‘solved’ the problem of generative AI in litigation, it could be argued that this was never its goal. With cases of not just self-represented litigants, but legal solicitors trusting its hallucinations, it was clear that there had to be some sort of framework to quickly address the use of Gen AI in legal practice. Its best achievement lies in how it has converted a vague professional anxiety into a regimented set of courtroom rules. It instructs lawyers as to what the court will tolerate, what it will scrutinise, and what it will not excuse. The FCA’s recent GPN-AI represents how this conversation has broadened from intervention by the courts to responsible management of Gen AI. This ultimately shows that SC Gen 23 was not a ‘knee-jerk reaction’, but a product of its time. It was an early attempt to impose discipline on a technology that legal practice could not ignore, but also could not trust blindly, especially with its imminent threat to miscarriage of justice.

VII REFERENCES

[1] Federal Court of Australia, Use of Generative Artificial Intelligence Practice Note (GPN-AI), 16 April 2026.

[2] Supreme Court of New South Wales, Practice Note SC Gen 23 – Use of Generative Artificial Intelligence, 21 November 2024 (commenced 3 February 2025) para 9B.

[3] Ibid para 4.

[4] Ibid para 6 (a-b).

[5] Ibid para 10-13.

[6] Ibid para 16.

[7] Practice Note SC Gen 23 (n 2) para 19-22.

[8]  Grace Robbie, ‘Barristers Weigh in on NSW Supreme Court’s AI Practice Note’ (5 December 2024) <https://www.lawyersweekly.com.au/biglaw/41107-barristers-weigh-in-on-nsw-supreme-court-s-ai-practice-note> [12].

[9] Ibid [33].

[10] Henry Ea, ‘NSW Supreme Court’s Gen AI Practice Note Sparks Debate’, Law Society Journal (16 December 2024) [2] <https://lsj.com.au/articles/nsw-supreme-courts-gen-ai-practice-note-sparks-debate/>.

[11] Ibid [3].

[12] Ibid [7].

[13] Centre for the Future of the Legal Profession, GenAI, Fake Law & Fallout: A review of the misuse of generative artificial intelligence in legal proceedings, Report (November 2025) 7.

[14] Dayal [2024] FedCFamC2F 1166.

[15] JNE v Minister for Immigration and Citizenship [2025] FedCFamC2G 1314.

[16] ‘NSW Supreme Court’s Gen AI Practice Note Sparks Debate’ (n 10).

[17] May v Costaras [2025] NSWCA 178.

[18] Bottrill v Graham & Anor (No 2) [2025] NSWDC 221 68.

[19] Michael Legg, Vicki McNamara and Armin Alimardani, ‘The Promise and the Peril of the Use of Generative Artificial Intelligence in Litigation’ (2025) 48(4) University of New South Wales Law Journal 34 <https://www.unsw.edu.au/content/dam/pdfs/law/unsw-law-journal/2020-2029/2025/48(4)-article3-Legg-et-al.pdf>.