The short answer
A law firm should not treat “AI” as one thing. Before client information is entered into any AI tool, the firm needs a deliberate decision about the exact service, account type, information, purpose, vendor handling, access permissions and human review.
The more useful question is not “Can we use AI?” It is: what exactly would be entered, into which approved service, by whom, for what purpose, and what happens to the result?
The Office of the Privacy Commissioner of Canada explains that organizations using AI still need to address privacy principles such as accountability, safeguards, transparency and limits on collection, use and disclosure. For Ontario law firms, that operational question sits alongside professional obligations and the firm’s own file, client and practice-management procedures.
Why “remove the name” is not a complete control
Removing a client’s name may not remove the ability to identify the person or matter. Dates, facts, property details, transaction amounts, opposing parties, communications, document wording and combinations of otherwise ordinary details can still be sensitive. A partial redaction is not a substitute for a firm decision about whether the tool and use are appropriate.
Start with the strictest practical default: staff do not enter client, confidential, privileged, financial, identity or security information into public or unapproved AI tools. Any exception should use an approved service and an approved workflow.
Five questions before AI receives client-related information
What is the exact tool and plan?
“ChatGPT,” “Copilot” and “Gemini” are families of products, not one uniform service. A free personal account, paid individual plan, business workspace and enterprise service can have different administration, contractual, retention and privacy features. The firm should identify the exact plan before approving a use.
Who owns and administers the account?
The account should be visible to the firm, protected with MFA, assigned to an accountable owner and removable when a user leaves. A tool used through a lawyer’s or employee’s personal account is difficult to supervise, export, investigate or recover.
What information is actually necessary?
Define a business purpose narrowly. An approved use might involve public legal updates, generic drafting structure or synthetic examples. A use involving a client’s facts, communications, identification, banking data, signed material or full file content requires a much higher level of review and may be unsuitable for the tool.
Does the tool connect to firm data?
Embedded assistants and connectors can access information differently from a standalone prompt. When a tool is grounded in Microsoft 365, a document management system, email, shared drive or other repository, review permissions and connections—not simply the prompt text.
Who verifies the result?
AI output may be inaccurate, incomplete or fabricated. The responsible licensee or staff member must verify the output before it is relied upon, sent externally or added to a file. The firm should define review expectations for each use case.
Public chatbot, approved workspace or Microsoft Copilot?
These are not interchangeable. A public tool may be appropriate only for public or synthetic material. A firm-controlled business service may offer better administration and contractual protections, but that does not remove the need for information rules. Microsoft Copilot may operate within the user’s existing access permissions, which makes SharePoint, OneDrive, Teams and shared-mailbox hygiene central to the decision.
Read the related guides on confidential files and Copilot and Microsoft Copilot in a law-firm environment.
If something was entered into the wrong tool
Do not treat it as a private embarrassment. The firm needs a simple escalation path. Record the tool and account involved, the information entered, whether it was uploaded or connected, who had access, what action has already been taken and whether the matter requires technical, privacy, professional or legal review. Quick facts make it easier for the firm to make informed decisions.
A practical interim rule
While the firm completes a fuller AI review, a short rule can state:
- Only identified firm-controlled accounts may be used for approved AI work.
- Client, confidential, privileged, financial, identity and security information is prohibited by default unless a documented workflow is approved.
- AI output must be reviewed by a competent person before use.
- New AI tools, plug-ins, connectors and browser extensions require approval.
- Possible disclosure, inaccurate output or account compromise must be reported promptly.
Related CyberTECT guidance
Authoritative sources and further guidance
- Law Society of Ontario — Using technology
- Law Society of Ontario — Licensee use of generative AI
- Canadian privacy regulators — Principles for responsible generative AI
General information only. This article does not provide legal, privacy or professional-conduct advice, determine privilege or confidentiality obligations, or certify Law Society compliance.
CyberTECT resources provide general operational guidance. They do not replace advice specific to your legal, regulatory, contractual or technical circumstances.