In most law offices, the question is no longer whether AI will arrive. A legal assistant has already used a chatbot to tidy up a letter, and a junior lawyer has already asked one to explain a procedure. The real question is whether it happens with rules, or without them.
In a legal office, the stakes are higher than elsewhere: confidentiality, privilege, professional duties and, sometimes, a judge reading the result. Here is a practical starting point.
What the rules already say
You do not have to start from zero. In Alberta, two documents already set the tone:
- The Law Society of Alberta's Generative AI Playbook (late 2023) reminds lawyers of their duty of technological competence, names the main risks (confidentiality, invented content, bias) and recommends verifying every output and using AI to support legal judgment, never to replace it.
- The October 2023 notice of Alberta's three courts (Court of Justice, Court of King's Bench and Court of Appeal), Ensuring the Integrity of Court Submissions when Using Large Language Models, asks parties to rely on authoritative sources such as official court websites or CanLII, and to keep a "human in the loop" who verifies what AI produces.
On top of that, the personal information of your clients is protected by privacy laws such as Alberta's PIPA. These texts evolve, so always check the current version with the Law Society and the courts.
What you can do safely
With an approved tool and clear rules, AI saves real time on work that does not require legal judgment:
- First drafts of routine letters and emails, written from anonymized facts and always reviewed by a lawyer;
- Intake triage: incoming requests summarized and sorted, with a checklist of missing documents;
- Plain-language explanations of a process for clients, reviewed before sending;
- Internal meeting notes turned into minutes and task lists;
- Deadline reminders and follow-ups on unanswered messages;
- Translation of non-confidential material between English and French, checked by a fluent reader.
Five things never to do
1. Paste client information into a public AI tool. Names, file details, privileged communications or anything that identifies a client stay out of consumer chatbots. If a task needs real client data, it needs an approved tool with verified settings, or no AI at all.
2. Cite anything AI produced without checking it. AI can invent cases, quotes and section numbers that look perfectly real. Canadian courts have already sanctioned lawyers over fictitious AI-generated citations. Every authority is checked in an official source before it goes anywhere.
3. Let AI send anything on its own. No message to a client, opposing counsel or the court leaves the office without a lawyer reading it. Automations prepare; people decide.
4. Use a tool whose terms you have not read. Where is the data stored? Is it used to train the model? Can you delete it? If nobody in the office can answer, the tool is not approved yet.
5. Leave it to each person to decide. The riskiest situation is not a bad tool, it is ten people using AI ten different ways. Write the rules down, once, for everyone.
A simple test before using AI on a file: "Would I be comfortable explaining this use to the client, and to the Law Society?" If not, don't.
A one-afternoon starting point
- List the AI tools people already use, without judgment.
- Classify your information: public, internal, and confidential or privileged.
- Choose one approved tool, check its settings, and make it the official one.
- Name one person who answers questions and receives incident reports.
- Write a one-page rule and walk your team through it.
This article is general information, not legal advice. For questions about your professional obligations, refer to the Law Society of Alberta and your own advisors.
To get started, download our free AI use policy template, which includes a three-level classification ready to adapt. Or see how AI-Ready 30 sets up rules, training and safe automations for legal and professional offices, in English or French.