Yes — lawyers in every US state and Canadian province may use ChatGPT and similar generative AI tools, provided they satisfy the professional obligations that already govern all technology use: competence, confidentiality, supervision, candor to the tribunal, and reasonable fees. No US bar or Canadian law society has banned generative AI. What regulators have done — most notably the American Bar Association in Formal Opinion 512 (July 2024) and the Law Society of Ontario in its guidance on licensee use of generative AI — is confirm that the existing rules apply with full force, and that a lawyer who submits unverified AI output is personally responsible for every word of it.
That framing matters because the question most lawyers are really asking is not "am I allowed" but "how do I use this without ending up in a sanctions decision." The cautionary cases are real: in Mata v. Avianca, Inc. (S.D.N.Y. 2023), two New York lawyers were sanctioned $5,000 under Rule 11 after filing a brief containing six fabricated cases generated by ChatGPT — complete with invented quotations and internal citations. The Second Circuit referred counsel for discipline in Park v. Kim (2d Cir. 2024) for the same failure. In Canada, Zhang v. Chen (B.C.S.C. 2024) saw a lawyer ordered to personally pay costs after citing two non-existent cases hallucinated by ChatGPT, and Ontario courts have since confronted similar fake-citation filings. Every one of these lawyers could have avoided the outcome with a ten-minute verification pass.
The adoption numbers explain why the question has become unavoidable. Legal-industry surveys from 2024 through 2026 — including the ABA's technology survey and the annual reports from major practice-management vendors — show generative AI use among lawyers climbing from a niche experiment to a substantial minority and, in larger firms, a majority of practitioners using AI for at least some tasks. Clients are moving even faster: corporate legal departments now routinely ask outside counsel about their AI capabilities in RFPs, and consumer clients arrive at consultations having already asked ChatGPT about their legal problem. A firm that bans AI outright is not avoiding the issue; it is choosing to compete against firms that draft, summarize, and prepare faster — while its own associates quietly use personal accounts without policy, training, or confidentiality controls, which is the worst possible configuration.
So the honest, definitive answer is: use it, but treat it like a brilliant first-year associate with no access to a law library and a tendency to invent authority — everything it produces gets checked before it leaves your desk. This guide covers the ethics guardrails, the tasks where ChatGPT genuinely earns its keep, the tasks it should never touch, and a practical workflow your firm can adopt this week. It anchors our wider AI in Legal Practice library.
Confidentiality is the first and hardest guardrail. Under ABA Model Rule 1.6 — and its analogue in every state — a lawyer must make reasonable efforts to prevent unauthorized disclosure of information relating to the representation. In Canada, Rule 3.3-1 of the Federation of Law Societies' Model Code imposes a duty of confidentiality that covers all information acquired in the professional relationship. Consumer ChatGPT, by default, may retain conversations and use them to improve OpenAI's models. Pasting a client's name, facts, or documents into a default consumer account is a disclosure to a third party whose data practices you do not control.
ABA Formal Opinion 512 addresses this squarely: before inputting information relating to a representation into a generative AI tool that may retain or train on it, a lawyer should obtain the client's informed consent — and boilerplate consent buried in an engagement letter is not enough for self-learning consumer tools. The practical hierarchy most firms adopt:
Remember also that prompts are records. What your lawyers type into an AI tool can be preserved by the vendor, produced in vendor litigation, exposed in a breach, or — in at least one high-profile US case — subjected to court-ordered preservation that overrode the vendor's ordinary deletion practices. Treat every prompt as a document that could one day be read aloud by someone adverse to your client: that single habit does more for confidentiality discipline than any policy memo. It also argues for centralizing AI use on firm-managed accounts, where administrators can see usage, enforce settings, and answer a client's or regulator's questions about exactly what was shared and under what terms — questions that are unanswerable when work happens on personal logins.
Canadian counsel should also check their law society's cloud and outsourcing guidance — the Law Society of British Columbia and Law Society of Ontario both direct lawyers to understand where data is stored, who can access it, and whether it is used for training, before any confidential use. The full analysis, including vendor questionnaires and engagement-letter language, is in our companion guide to client confidentiality and AI tools.
ChatGPT is a language model, not a database. It predicts plausible text; it does not look anything up. That is why it fabricates case citations with such confidence — a Stanford study of general-purpose chatbots found hallucination rates between 58% and 82% on legal queries, and even purpose-built legal AI research tools produced incorrect or unsupported answers a meaningful fraction of the time. The duty of competence (ABA Model Rule 1.1, whose Comment 8 on technological competence has been adopted in some form by 40 US states, and FLSC Model Code Rule 3.1-2 in Canada) requires understanding this limitation. The duty of candor to the tribunal (Model Rule 3.3) makes submitting a fabricated citation a disciplinary matter regardless of intent.
The verification protocol is not complicated: every case ChatGPT names gets pulled in Westlaw, Lexis, or CanLII; every quotation gets confirmed against the actual reporter text; every proposition gets checked through a citator (KeyCite or Shepard's in the US; CanLII's noteup in Canada) to confirm it is still good law. If a cited case cannot be found in a real database, it does not exist. Courts have made clear that "ChatGPT told me it was real" is not a defense — the Mata court noted the lawyers even asked ChatGPT to confirm the cases were genuine, and it obligingly lied. We break the failure mode and the full protocol down in AI hallucinations in legal research.
Build the check into the file, not the lawyer's memory. Firms that have institutionalized verification use a short certification line in the document history — "authorities verified in CanLII/Westlaw on [date] by [initials]" — and make it a condition of filing, the same way a second signature is a condition of releasing trust funds. The cost is minutes; the payoff is that when a judge, client, or insurer asks how your firm prevents fabricated citations, you have a documented answer rather than an assurance. Malpractice insurers on both sides of the border now ask exactly that question on renewal applications, and a written verification protocol is quickly becoming the difference between a routine renewal and an uncomfortable conversation.
Verification also extends beyond citations. AI-drafted contract language gets read clause by clause against the deal terms. AI summaries of documents get spot-checked against the source. AI-suggested deadlines get confirmed against the actual rules of court. The time this takes is real — but it is a fraction of the time the AI saved, and infinitely less than the cost of a sanctions motion or a Law Society complaint.
Generative AI is strongest where the output is a first draft a lawyer will rework, and weakest where the output must be authoritative on its own. Tasks that fit well:
Tasks that do not fit — at least not with a general-purpose chatbot:
Practice-area texture helps make the line concrete. A family lawyer can safely ask ChatGPT to restructure a parenting-plan letter for a distressed client's reading level, but should never trust its statement of the current spousal support advisory ranges. A litigator can use it to generate a first-cut cross-examination outline from a transcript she uploads to an enterprise tool, but not to tell her what the limitation period is for the claim. A corporate lawyer can have it produce a board-minutes skeleton in seconds, but the securities-law compliance language comes from counsel, not the model. In every example the pattern is identical: AI supplies structure, speed, and prose; the lawyer supplies the law. Where a task requires the model to know current, jurisdiction-specific law on its own, the task is out of scope for a general chatbot — that work belongs in grounded legal research tools, and even there it gets verified.
Prompting technique deserves a sentence because it changes output quality more than tool choice. The reliable pattern for legal drafting is context-role-constraints: supply the relevant facts and any governing document, state the role ("outside counsel drafting for a sophisticated commercial client"), and constrain the output (jurisdiction, length, tone, provisions that must appear, propositions that must not be asserted without a source). Lawyers who prompt this way get drafts that need editing; lawyers who type one-line requests get generic text that needs rewriting — and then conclude, wrongly, that the tool is useless. Prompt templates for the firm's ten most common tasks are a one-afternoon investment that standardizes quality across the whole team.
Three further rule clusters shape everyday use. First, supervision: Model Rules 5.1 and 5.3 make partners and supervising lawyers responsible for ensuring that both subordinate lawyers and non-lawyer assistance — which regulators now read to include AI tools — are used in ways compatible with professional obligations. Practically, that means a written firm AI policy: approved tools, prohibited inputs, mandatory verification, and training. A firm whose associate files a hallucinated brief cannot shrug; the supervising partner owns the failure too.
Second, fees: under Model Rule 1.5 a fee must be reasonable, and ABA Op. 512 confirms you may not bill hourly time you did not spend — if AI turns a six-hour draft into forty minutes, you bill the forty minutes (plus your verification time), or move the work to flat fees that price the value rather than the hours. Passing AI subscription costs to clients as a disbursement requires disclosure and must reflect actual cost. The economics deserve their own treatment: see billing ethics for AI-assisted work.
Third, court disclosure: a growing set of US federal judges' standing orders and Canadian practice directions (the Federal Court of Canada, Manitoba's Court of King's Bench, Alberta's courts, and others) require certification or disclosure when generative AI contributed to a filing. Before filing anywhere, check the judge's standing orders and the court's practice directions — our guide to court rules on AI in filings catalogues the landscape.
Here is the adoption sequence that works for small and mid-sized firms in both countries:
Measure the results honestly, because the honest numbers drive good decisions everywhere else. Track minutes saved per task category, verification time as a share of total, error catches in review, and — critically — which tasks lawyers stopped using the tool for after trying it, since abandonment data is the fastest map of where the technology actually underdelivers. Firms that measure typically find the savings concentrate in a handful of document types, which tells them exactly what to productize on flat fees (a decision with its own ethics dimension, covered in the billing guide above), and that verification consumes 20–40% of the gross savings — a number worth knowing before promising clients or partners anything. Culture matters as much as policy: lawyers hide AI use in firms that treat it as cheating and disclose it in firms that treat it as a supervised skill, and only the second kind of firm can actually manage the risk.
Used this way, ChatGPT is what the sanctions decisions say it can be when handled properly: a legitimate drafting and thinking aid that returns hours to the lawyers who supervise it. Firms adopting AI internally usually discover the flip side too — their prospective clients are now asking ChatGPT which lawyer to hire. That marketing question is covered in our ChatGPT for law firms hub, and you can test your own firm's AI visibility in two minutes with the free AI Visibility Checker. For a firm-wide plan covering both sides, book a strategy call with LexScale.ai.
LexScale.ai helps law firms across Canada and the United States adopt AI for growth — from client-facing intake and content systems to the visibility that puts your firm inside AI answers.
Book a Free Strategy Call →The numbers behind these trends are collected in our law firm AI adoption statistics roundup — useful for citations and partner buy-in.
Unfamiliar with a term here? Our plain-English legal AI glossary defines the concepts firms encounter most.
This article is general information, not legal or ethics advice. Professional-conduct rules on AI are evolving and vary by jurisdiction — always verify current requirements with your state bar, law society, or regulator before adopting any AI workflow.
Related Articles
Client Confidentiality and AI Tools for Lawyers · Court Rules on AI in Filings: US & Canada Guide · Billing Ethics for AI-Assisted Legal Work · AI Contract Drafting for Lawyers: What Works · AI Document Review & Discovery: From TAR to LLMs · AI Hallucinations in Legal Research: Risks & Fixes
Ready to grow your firm with AI?