Client consent for
AI-assisted legal work.

The Law Society of BC’s guidance on AI is clear on the shape of the thing: lawyers remain responsible for competence, confidentiality, and the accuracy of AI-assisted work. No regulator approves specific products, and any vendor implying endorsement is telling you something no vendor can say. What a careful firm can do is work through the questions below, and expect specific answers, from any vendor including us.

The cautionary case every BC family lawyer knows: in Zhang v. Chen (2024 BCSC 285), counsel who filed submissions containing citations an AI tool had invented was held personally responsible for the costs her fabricated citations forced on the other side. The court declined special costs, finding no intent to deceive, which makes the lesson sharper: you do not need bad faith to end up personally paying for unverified AI output. The court called disclosure of AI use prudent, not mandatory.

1. What exactly is processed, and where? Which documents leave the firm’s systems, to which providers, hosted in which countries. “Canadian data residency” claims deserve a follow-up: residency of what, exactly? Storage and processing are different questions, and an honest vendor will answer them separately.

2. Is client information used to train anything? Get the contractual answer, not the marketing answer.

3. How long is anything retained by each provider? Named retention periods beat “industry standard.”

4. Is your firm’s data isolated from other firms’? Shared platform behind a login, or a system of the firm’s own: the architecture matters to the confidentiality analysis.

5. What does the client hear, and when? Whether and how AI use is disclosed to the client, what the retainer or engagement letter says, and what a written consent covers. A workflow that produces a record beats an intention.

6. What does the record prove? If the Law Society asks how the work was supervised, can the firm show when the professional-responsibility checks happened, what the AI produced, and where the lawyer decided?

7. How are citations and facts verified, and by whom? What is checked by code, what is checked by the AI, and what remains the lawyer’s own currentness and treatment review. The honest answer always ends with the lawyer.

8. What does the tool refuse to do? A system with wired-in limits, on sending, filing, signing, and guessing, is making the supervision analysis easier. A system that can do anything is making it harder.

CaseClarity ships with a written informed-client-consent workflow, engagement-letter language covering the AI processing, and disclosure handling, and the matter record shows the checks happened before the AI touched the file. Our own answers to every question above are on the security page, including the one vendors usually skip: where processing actually happens. The limits are on their own page: what it never does.