Supervising AI in a law firm: rule 37 in practice.
Under s 34 of the Legal Profession Uniform Law, principals are responsible for the legal services their practice provides, and must take reasonable steps to ensure everyone in the practice meets their professional obligations. Rule 37 adds the day-to-day duty of reasonable supervision. The Law Society of NSW guide applies both to AI directly, and the December 2024 joint statement expects firms to set out how they will continuously and actively supervise the use of AI tools by junior and support staff.
General information only. This guide summarises published regulatory guidance and is not legal advice. Read the primary sources linked throughout, and take professional advice on your firm’s specific position.
Why AI supervision is harder than it sounds
The guide makes an observation most firm policies miss: exercising reasonable supervision over AI use requires critical evaluation of the accuracy and completeness of the tool’s outputs, and a junior practitioner may lack the experience to make that evaluation. Supervision of AI is therefore a senior person’s job twice over: reviewing the output, and deciding which tasks are appropriate for AI assistance at all. The court finding discussed in our citations guide, where a supervisor’s failure contributed to a junior’s error, shows the principle applied.
What the regulators expect a firm to be able to show
The statement expects firms to decide and record which AI tools are in use in the practice, rather than discovering the answer after a problem.
Approved users, approved purposes, and approved categories of information, with higher-risk tasks reserved or prohibited.
Documents containing AI-generated content are reviewed for accuracy and verified by a qualified person before they are settled.
Continuous and active supervision of junior and support staff, which means someone senior can describe how AI-assisted work is checked, on real matters.
When and how AI was used in a matter, so the firm can answer a client, a court or a regulator without reconstruction.
The shadow AI problem
The hardest supervision problem is the tool nobody approved: staff using personal accounts on public chatbots because they are convenient. Prohibition alone tends to drive the behaviour underground. Firms that fare best pair a clear rule about public tools with an approved alternative that is genuinely useful, so the easy path and the compliant path are the same path. That is a supervision insight before it is a product one, though it is also the practical argument for giving staff a capable internal AI: one approved system, on the firm’s hardware, where usage is visible to the practice rather than scattered across personal accounts.
Where deployment posture fits
Supervision obligations are indifferent to architecture; a principal who fails to review work breaches rule 37 with any tool. What architecture changes is observability. A single on-site system gives the practice one place where AI-assisted work happens, which makes the statement’s expectations, who uses AI, for what, with what review, answerable in fact rather than on paper.
Common questions
Is a principal responsible if a junior misuses AI?
How do firms deal with staff using ChatGPT on personal accounts?
What does 'reasonable supervision' of AI look like?
See Dominion answering from your own documents.
A demonstration takes 30 minutes, uses no client data, and comes with no obligation.
Book a demo