Trust, but verify: supervising junior lawyers using AI

Supervised legal practice typically involves experienced practitioners instructing, mentoring and assuming responsibility for a less-experienced practitioner’s work.

But supervision has a wide meaning. Any solicitor can be a supervisee unless they have carriage of the matter, and holders of unrestricted practising certificates who commit disciplinary infractions may find a supervision condition reimposed.

Supervised legal practice

Here, we consider new entrants to the profession, that is, the requirement that newly admitted lawyers must engage in 18 to 24 months of supervised legal practice before ‘Condition 2’ can be removed from their practising certificate. This requirement is a modern, statutory version of the apprenticeship model of legal training. Despite its longevity, the supervision of new lawyers continues to be a source of dissatisfaction for some supervisees (see Early career lawyers’ experiences of supervised legal practice: 2023 survey | VLSB+C). Relatively speaking, however, very few cases result in disciplinary or other legal action. Those that do tend to involve extreme fact circumstances, where poor supervision is one of a multitude of other ethical and practice issues. For instance, in Victorian Legal Services Commissioner v Knight (Legal Practice) [2024] VCAT 887, the principal made the seemingly appropriate concession ‘that he was probably not providing appropriate supervision to junior lawyers when he was coming down off cocaine’ (at [168]). Another category of cases involve junior practitioners committing ethical breaches, which may call into question the adequacy of the supervision they have received (see, e.g., Legal Services Commissioner v Lee [2026] QCAT 34); or appearing in court without proper supervision (Titus & Ksenia [2026] FedCFamC1F 104; Council of the Law Society of New South Wales v Judah [2022] NSWCATOD 89).

The misuse of generative AI (‘GenAI’), unwittingly resulting in the putting of false authorities before a court, has raised the issue of supervision in less extreme, even mundane, circumstances. A look at these cases illustrates both some poor work practices which typically would not come to light in a court setting; and the additional issues that AI tools can generate, including undisclosed AI use, or overreliance, which may contravene firm policy and compromise final work product. In short, GenAI misuse cases can tell us something about supervision more generally, as well as the specific risks of AI tools.

Supervision goes wrong – generative AI examples

GenAI cases are useful as examples of ‘run of the mill’ poor supervision, rather than the extreme examples coming before courts or disciplinary tribunals. Recent cases involving the misuse of generative AI by a supervisee provide a window into problematic work practices and pitfalls, including the challenges of remote supervision and misplaced trust. For instance, in Murray on behalf of the Wamba Wemba Native Title Claim Group v State of Victoria [2025] FCA 731 (‘Wamba Wemba’) (discussed in LSJ Online, November 2025, by Farazi), a junior solicitor, working remotely, used Google Scholar and GenAI to generate citations for court documents. The citations were primarily references to ‘anthropological and historical reports’ relevant to the applicants’ native title claim. The errors led to the filing of misleading evidence. Surprisingly, the junior solicitor deposed to preparing the footnotes without access to either ‘physical or electronic copies of the footnoted documents’ (at [6]). Explaining supervisory practices at the firm, the supervising solicitor referred to work being ‘performed collaboratively between team members’ (at [9]). However, the crux of the issue was that no one had checked the junior solicitor’s work. More than that, there appeared to be no policy, guidance or even accepted practice as to how remote working should be undertaken, how the checking of work should occur in a team, and certainly not how those two elements should be combined.

In the English case of Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin) (‘Ayinde’), the High Court considered two separate instances where lawyers had put false authorities and statements of law before the Court. In Ayinde, the barrister responsible for the fake material was a pupil barrister who claimed to have received no supervision in her second six months (the practising period) of pupillage, including none of her written work being checked (at [51]). This was denied by the Director of her chambers in an email to the Court. Nevertheless, the Court concluded: ‘there are questions raised as to potential failings on the part of those who had responsibility for training [the barrister], for supervising her, for “signing off” her pupillage, for allocating work to her, and for marketing her services’ (at [69]). The Court also referred the instructing solicitor to the legal regulator to investigate further ‘the steps he took to satisfy himself that [the barrister] had sufficient experience or was competent to undertake the work she had been instructed to do by him’ (at [72]). The paralegal involved was found to be blameless (at [63]).

Similar ‘chain of command’-type issues are raised by the Australian family law case Mertz & Mertz (No 3) [2025] FedCFamC1A 222 (‘Mertz’), involving a paralegal, solicitor and two counsel. Counsel had settled the applicant’s summary of argument and list of authorities submitted to the Court, which included hallucinated material. Both counsel ‘provided detailed explanations of the circumstances (including that they had not used AI themselves), accepted responsibility, apologised to the Court and provided reassurance that the error would not be repeated’ (at [5]). However, the solicitor’s explanation was less fulsome. She advised that her paralegal had used AI to alphabetise a list of references but had also then inserted some of these references into footnotes. The Court said: ‘The submissions of [the solicitor] did not identify which AI program or programs had been used. Her submissions did not identify what, if any, training, supervision or guidance the paralegal had been given in relation to the use of AI’ (at [8]). Presumably, this would be appropriately investigated by the relevant disciplinary body to whom referral was made. The Court noted the solicitor advised she had ‘terminated’ the paralegal’s services (at [8]).

The Court in Mertz also noted that while a second version of the documents had purportedly corrected the AI-generated errors, this was incomplete, as some errors remained. The solicitor was ordered to pay $10,000 towards the respondent’s wasted costs arising from the need to correct AI-generated errors.

These instances of combined supervision and AI failures illustrate some useful points. One concerns the importance of workflow process and structured guidance. As Ritchie J observed in the first iteration of Ayinde v the London Borough of Haringey [2025] EWHC 1040 (Admin), ‘it is the responsibility of the legal team… to see that the statement of facts and grounds are correct’ (at [64], emphasis added). Who is responsible for checking work? And saliently, in the ‘AI cases’, what exactly needs to be checked? Was case law obtained through an authorised database, or from a chatbot? Was GenAI used to summarise a document? While interrogating a colleague about their work methods might be awkward, it is vital that supervisors understand the work practices of those they are supervising. Alarm bells might sound: is a response unaccountably wrong (e.g., the defensive letter written by the barrister in Ayinde), or was work produced too quickly to have been done without AI?

This leads to the second point – that collaborative working requires trust, which AI complicates. In several cases of AI misuse, errors have been missed by multiple people involved in a matter: solicitors, counsel and even the opposing side (e.g. DPP v GR [2025] VSC 490). AI use can make a person appear more capable when it comes to traditional measures of gauging this, such as fluent writing, or completing research tasks quickly. But this could be masking a lack of skills or adherence to agreed and appropriate work methods.

The final point concerns power imbalance, which inevitably exists between supervisor and supervisee. Supervisors cannot prevent supervisees from making mistakes or having ethical lapses, nor control how the supervisee reacts to the mistake (see Professor Richard Moorhead’s comment on Ayinde in this regard). But any supervisor with carriage of a matter has responsibility to oversee the work on that matter. Attempting to shift blame to a more junior person, especially if that person is not admitted as a practising lawyer, is unfair and could taint that person’s career before it starts (see McNeil & Rydell [2026] FedCFamC2F 149 at [8]). As with all disciplinary proceedings, legal professional privilege may at times obscure how a mistake unfolded. For instance, initial findings in Ayinde blamed the paralegal involved, though this was fortunately corrected by the High Court.

AI in legal supervision – what can supervisors do?

A recent report by the Centre for the Future of the Legal Profession (‘CFLP’), Making Supervision Work for Everyone, highlighted AI-assisted legal work as a ‘distinctive contemporary supervision risk’ with concerns around work methods, final work product and how supervisors can guide the process. Law firms must adapt their work practices to AI, whether or not principals endorse its use. This includes creating a culture of transparency and trust, that is, where everyone in the firm has transparency regarding AI and where it can and cannot be used; and where individual supervisors and supervisees trust one another to use AI responsibly and disclose its use. Good supervision practices can contribute to this.

1. Develop and maintain a clear supervision plan, implement an AI policy and connect both to ethical practice

Wamba Wemba, Ayinde and Mertz all illustrate the importance of transparency in work practices and ‘speaking up’, whether between supervisor and those being supervised, or between colleagues. Effective supervision cannot be informal or assumed (Giddings and McNamara, 2014), especially in hybrid or remote environments which are increasingly common in law firms.

Structured supervision plans should explicitly reference a firm’s policy and processes surrounding GenAI use, including clarifying when AI may be used, how outputs must be reviewed and who bears responsibility for the final work. Drawing new lawyers’ attention to the Law Society’s Solicitor’s Guide to Responsible Use of Artificial Intelligence and instilling this into the firm’s policy helps to situate GenAI within the profession’s ethical framework. Ensuring there are protocols to identify, verify, correct and document AI-generated content alongside additional checks to prevent errors or hallucinations from passing unnoticed, is vital.

It is important that a supervision plan and AI policy do not operate in silos and are harmonised to ensure that supervisors know when AI may have been used and can review work accordingly. This reinforces the fact that final responsibility for work product lies with supervisors, consistent with the Uniform Law and professional conduct rules. AI governance must be embedded through supervision plans and policies to form part of the firm culture, rather than being an optional add-on. Normalising transparent GenAI use can create an environment of ongoing internal education regarding its ethical and practical limits. Supervisors might also use GenAI beneficially in their own work, or even as part of supervision, such as to give first-round feedback.

2. New lawyers may find their work challenging – ensure that they don’t turn to AI in secret

In many instances of AI misuse, lawyers appear to have simply misunderstood how the AI tool works, the types of mistakes that it can make and the need to verify (e.g. Dayal [2024] FedCFamC2F 1166). Traditionally, work done by paralegals, graduates and newly-admitted lawyers might be thought of as ‘low-risk’, such as completing footnotes or lists of authorities. However, using AI without appropriate care creates the possibility that mundane tasks can be catapulted into ‘misleading the court’ territory. Supervisors also need to be aware that supervisees may lack the knowledge and experience to easily verify AI outputs. As they are learning both to practice law and, most likely, how to use AI tools, instruction and guidance may need to encompass both.

More widely, there are conflicting messages regarding the use of GenAI by lawyers. New lawyers face pressure to be efficient and technologically capable, whilst simultaneously experiencing fear of reputational damage given the stigma surrounding AI misuse (Bell & Rogers, 2025). Mistakes made by lawyers because of GenAI are often treated as arising from ignorance or a lack of understanding of the tools used, or laziness by trying to ‘cut corners’ in their work, and may be stigmatised both within an organisation and in the wider legal sector (e.g., Lawyerly; Schnitzer & DeGregory, 2023). These conditions may produce uncertainty, ethical anxiety and a tendency toward concealment for early-career lawyers. As well as a harmonised supervision plan and policy around AI, it is vital that legal organisations foster an environment in which lawyers can appropriately discuss any challenges in their work, especially those arising from time pressures.

Conclusion

AI, and new ways of working in general, make supervision increasingly important and emphasise the need to develop new graduates into lawyers with strong technical skills, critical thinking (including in relation to new technologies), and an awareness of their ethical obligations. The goal of supervision is to develop lawyers who are confident in their abilities and accountable for their final work product. Consequently, supervisors and legal organisations have the responsibility to continuously evolve their work practices to new ways of working, such as creating a culture of transparency, ethical use and trust around AI.

Dr Felicity Bell is a Senior Lecturer and Deputy Director of the Centre for the Future of the Legal Profession at UNSW Law and Justice. Vishwa Shah is an intern, also at the CFLP.