Using it is not the ethical question. What you bill, what you tell the client, what you tell the court and whether you checked — those are.
Key takeaways
Using AI is not, in itself, an ethical breach — Canadian law societies treat it as a competence and confidentiality question rather than a prohibition, and the Law Society of British Columbia’s stated goals include ensuring effective tools are in place to support appropriate AI use, reducing practice infractions related to AI, and increasing the public’s access to legal services from trusted, AI-based sources.
The genuinely awkward questions are narrower: what you bill, what you tell the client, what you tell the court, and whether you actually checked. Each has a defensible answer and an indefensible one.
Ontario’s Rules of Professional Conduct require that a fee be fair and reasonable and disclosed in a timely fashion. The commentary lists what "fair and reasonable" depends on, and it is a longer list than time: the time and effort required and spent, the difficulty and importance of the matter, whether special skill or service was required and provided, the amount or value involved, the results obtained, the lawyer’s experience and ability, any estimate given, and the client’s prior consent.
So the ethical problem is not that a drafting task took twenty minutes instead of two hours. It is billing two hours anyway. If your fee genuinely rests on judgement, results and skill, say so and price on that basis; the rule already contemplates it. If it rests on time, the time has to be real. The commentary is equally clear that the fiduciary relationship requires full disclosure in financial dealings and prohibits hidden fees.
There is no general rule requiring a lawyer to disclose which software they used, any more than which word processor. Two things change that. The first is confidentiality: if the client’s material is going to a third-party system, the analysis is about disclosure and consent, not about tooling. The Lawyers Indemnity Fund’s notice on generative AI puts the question directly — if you cannot redact client confidences before use, is client consent necessary?
The second is materiality. Where AI shapes the substance of advice rather than the speed of producing it, a client who later learns that from somewhere other than you will reasonably feel misled. Say it in the retainer, once, and stop having the conversation.
This is the one with a documented body count. The Lawyers Indemnity Fund’s notice describes a United States case, Mata v Avianca, in which counsel relied on ChatGPT for a submission without verifying the authorities and filed citations to non-existent cases — and, when challenged, asked the same tool to produce the full citation, which it duly fabricated. It is an American decision, but the lesson is jurisdiction-neutral.
Canadian courts are addressing it. The same notice records that some courts in Canada have begun requiring disclosure of the use of generative AI tools in submissions, and advises checking with any court, tribunal or authority on its AI policy before relying on the output. Directions differ by court and they change, so confirm the current one before you file — every time, on every court.
The Code of Professional Conduct for British Columbia and its Ontario equivalent both impose a duty of confidentiality that is wider than the evidentiary rule of privilege. Putting client material into a service you have not diligenced is a confidentiality decision whether or not you thought of it as one.
Where privilege will later be asserted, the stakes rise further — privilege claims in discovery turn on how material was handled, and solicitor-client privilege is not a thing to test experimentally.
Ontario’s competence commentary, amended in June 2022, expects a lawyer to develop an understanding of and ability to use technology relevant to their practice, and to understand its benefits and risks. Read plainly, that is not only a warning about using AI carelessly. It is also a reason not to ignore a tool that materially affects what your clients can afford — which is the access-to-justice half of the argument, and the half the profession is least comfortable discussing.
Neither half licenses the other. Enthusiasm does not excuse an unverified citation, and caution does not excuse charging a client for work you did not do.
Verify every authority, quotation, figure and date against the primary source. Keep confidential material out of any tool whose data terms you cannot describe. Bill on a basis you would be content to explain to the client and to a taxing officer. Say what you do in the retainer. Keep a short record on the file of what the tool did and who checked it. And read your court’s practice direction before you file.
Is it unethical to bill hourly for AI-assisted work?
Not inherently. It is unethical to bill for time you did not spend. If the tool made you faster, the client gets the benefit or the basis of the fee changes — and the rules already list skill, difficulty and results as legitimate factors.
Do we have to tell the court?
It depends on the court, and directions are changing. Some Canadian courts have begun requiring disclosure. Verify the current practice direction rather than relying on what was true last year.
Is refusing to use AI the safe position?
Safe from one risk, exposed to another. Competence includes understanding technology relevant to your practice, and a client who paid for forty hours of document review that a supervised tool could have narrowed in four may eventually ask why.
A 30-minute call is enough to tell you whether AI pays for itself here.