Using ChatGPT does not automatically destroy attorney-client privilege, but lawyers should not treat the answer as a simple yes or no. Privilege can depend on what information was shared, why it was shared, which system received it, what protections applied, and the law of the relevant jurisdiction.

In Canada, the more common term is solicitor-client privilege. The Law Society of Ontario explains that it protects confidential communications between a lawyer and client made for the purpose of seeking or giving legal advice, and that the privilege belongs to the client.

Why AI creates a privilege question

Generative AI often requires users to provide context. For legal work, that context can include client instructions, legal advice, draft opinions, case strategy, witness information, or communications between the lawyer and client.

The Canadian Bar Association warns that lawyers using generative AI should consider risks involving confidentiality and potential loss of privilege. Its guidance also recommends that lawyers reasonably ensure the security of an AI system before using it and consider whether the client should be informed about how generative AI will be used in the matter.

The issue is not simply that an AI provider is a third party. Lawyers already use third-party technology providers for email, document management, cloud storage, research, and other services. The relevant legal analysis can depend on the circumstances, including the purpose of the disclosure and the safeguards surrounding it.

That is why firms should avoid statements such as “ChatGPT always waives privilege” or “our enterprise account means privilege is protected.” Both are too absolute.

Separate privilege from confidentiality

Privilege and confidentiality overlap, but they cover different things.

The Law Society of Ontario explains that the duty of confidentiality is broader. It can cover all information concerning the business and affairs of a client acquired during the professional relationship. Solicitor-client privilege is narrower and generally concerns confidential lawyer-client communications made for the purpose of seeking or providing legal advice.

A prompt could therefore create a professional responsibility problem even when the content is not privileged. Client identities, commercial information, settlement positions, or factual records may still be confidential.

A strong AI policy should protect both categories rather than focusing only on documents labelled “privileged.”

Do not send privileged material simply because the task is convenient

Imagine a lawyer wants ChatGPT to improve the clarity of a draft email containing legal advice. Pasting the entire email may provide useful context, but the task may not require the client's name, specific transaction details, or the substance of the legal advice.

A safer approach is to remove matter-specific details and provide only the wording that needs assistance. For some tasks, a fictionalized version can accomplish the same goal.

Data minimization does not answer every privilege question, but it can prevent unnecessary disclosure when the sensitive material was never needed for the task.

Review the AI provider and the specific account

The security and contractual terms of the AI environment matter.

The Canadian Bar Association recommends due diligence when choosing AI tools, including review of confidentiality, privacy, security, contractual terms, integrations, and how input and output data are handled.

That review should include whether prompts are retained, whether customer content is used for model improvement, who can access submitted information, whether the firm controls retention, and whether integrations can reach client data.

An enterprise product with stronger contractual and technical protections may reduce certain risks compared with an unmanaged consumer account. It still does not create an automatic legal conclusion that every privileged communication can safely be submitted.

Client communication can also matter.

The Canadian Bar Association advises lawyers to consider whether clients know how generative AI is being used and whether meaningful consent is appropriate. It specifically identifies potential breaches of confidentiality and loss of privilege as risks that may need to be explained.

The American Bar Association's Formal Opinion 512 similarly says lawyers must consider duties of confidentiality and communication when using generative AI, and discusses circumstances where informed consent may be required before inputting representation-related information into some generative AI tools.

The exact requirement can differ by jurisdiction and use case. Firms should give lawyers a clear path to seek ethics or legal guidance instead of expecting a universal policy to resolve every privilege question.

Do not forget staff and client-side AI use

Privilege risk is not limited to partners typing into ChatGPT.

Articling students, assistants, clerks, paralegals, contractors, and other staff may all handle privileged communications. Firms should train them on which AI tools are approved and which types of information must stay out of prompts.

Clients can also use AI independently. A client might paste a lawyer's advice into a chatbot to ask for another explanation or upload litigation documents into an AI research service. That can create separate privilege questions outside the firm's direct control.

For particularly sensitive matters, lawyers may want to discuss appropriate AI use with clients when that risk is realistic.

Use a privilege check before submitting the prompt

Before using ChatGPT for client-related work, ask whether the material contains a confidential lawyer-client communication, whether the AI actually needs that information, whether the tool has been approved for the intended use, what contractual and technical safeguards apply, and whether client communication or consent is required.

If those questions cannot be answered confidently, the safer option is to remove the privileged information or obtain matter-specific guidance first.

Using ChatGPT does not produce an automatic privilege waiver in every situation, and using an enterprise AI product does not automatically guarantee privilege either. The outcome can depend on facts and jurisdiction.

For law firms, the practical approach is to minimize disclosure, use reviewed tools, understand vendor terms, communicate with clients when appropriate, and escalate privilege questions that depend on legal analysis. AI can assist legal work, but decisions about privileged information should remain deliberate.

Every safeguard discussed here shares a single constraint: once a prompt reaches an AI tool, it cannot be unsent. A privilege check that runs only in a lawyer's memory, or a policy that depends on someone pausing before pressing enter, will eventually miss the moment that matters. This is the principle Sanitized Ai is built on. The point of intervention has to be before submission, catching confidential lawyer-client communications and matter-specific details and removing them from the prompt while the wording that actually needs help still gets through.

This quarter, take one recurring AI task your lawyers already perform, such as tightening the language of advice emails, and map exactly what information that task requires versus what tends to get pasted in out of convenience. The gap between those two is your privilege exposure, and closing it is more concrete than any firm-wide statement about ChatGPT. If you want to see how disclosure can be minimized before a prompt is ever submitted, request a demo.