Think Before You Prompt: When Clients Share Confidential Information with AI

Clients may turn to artificial intelligence tools to review contracts, summarize documents, explain legal advice, or assess disputes. For example, a client might upload an agreement before sending it to counsel, paste a lawyer’s email into a chatbot for a simpler explanation, or ask an AI platform to evaluate the strengths and weaknesses of a case.

The risk is easy to overlook. Entering information into an AI service may involve transmitting it to an external provider. If that information is confidential, privileged, or personal, or if it concerns existing or anticipated litigation, an ordinary-looking prompt may have consequences extending beyond the immediate response.

Different AI Tools Offer Different Protections

AI platforms do not all handle information in the same way. Their practices may vary depending on the provider, product, account type, settings, and connected services.

A paid subscription, password-protected account, or enterprise account does not, by itself, establish that a tool is appropriate for sensitive information. Before using an AI service, clients should consider:

• whether prompts and uploaded documents are retained;
• whether content may be used to train or improve AI models;
• who may access the information; and
• what contractual confidentiality and security protections apply.

Using AI to summarize public information or ask a general question without providing sensitive facts presents different considerations from uploading legal advice, evidence, personal information, or internal business records.

Confidentiality and Privilege Are Not the Same

Information does not need to be legally privileged to require protection. Financial records, customer information, trade secrets, business strategies, employment records, and internal investigations may all be confidential. Uploading these materials may also engage contractual, privacy, regulatory, or other legal obligations.

Solicitor-client privilege generally protects confidential communications between a lawyer and client made for the purpose of seeking or providing legal advice.¹ Litigation privilege is distinct and may protect materials created or gathered for the dominant purpose of preparing for existing or reasonably anticipated litigation.²

A client may be tempted to paste an email from counsel into an AI platform to obtain a summary, plain-language explanation, or second opinion. Doing so may call into question whether the confidentiality necessary to preserve solicitor-client privilege was maintained. The consequences will depend on the circumstances, including the service used and the protections in place.

Canadian courts have not yet provided clear guidance on how privilege applies when clients enter protected information into consumer-facing AI tools.³

A recent American decision illustrates the practical risk. In United States v Heppner, a represented criminal defendant independently used the consumer version of Claude to develop defence materials without counsel’s direction. The court concluded that the resulting documents were protected by neither attorney-client privilege nor American work-product protection. Among other considerations, the AI platform was not the defendant’s lawyer, and the applicable privacy terms undermined the claim that the interaction was confidential.⁴

American work-product protection differs from Canadian litigation privilege, and Heppner does not determine how a Canadian court would treat similar materials. It nevertheless demonstrates why clients should not assume that AI-assisted work will receive the same protection as confidential communications with counsel.

AI Chats Can Create Additional Records

An AI interaction may create additional electronic records, including prompts, follow-up questions, summaries, assessments of evidence, proposed arguments, and saved chat histories.

If litigation is underway or reasonably anticipated, relevant AI records may need to be preserved. Relevant prompts and outputs may also be subject to production if they are not privileged and otherwise meet the applicable discovery rules.

These concerns are particularly relevant when a client uses AI to analyze evidence, reconstruct events, assist with testimony preparation, formulate settlement positions, or develop litigation strategy. Clients should consult counsel before using AI for these purposes.

Practical Takeaways

• Do not enter privileged or sensitive information without appropriate approval or advice. Businesses should use only AI systems approved for the relevant category of information. Individual clients should consult their lawyer before entering information connected to a legal matter.
• Understand the particular service being used. Do not assume that a paid, password-protected, or organization-managed account is appropriate for confidential information. Review the applicable retention, training, access, confidentiality, and security terms.
• Exercise caution in litigation. Consult counsel before using AI to analyze evidence, assist with testimony preparation, evaluate legal advice, or develop strategy.
• If sensitive information has already been entered, contact counsel promptly. Where litigation is underway or reasonably anticipated, preserve the interaction and obtain advice before deleting or modifying any related records.

This article provides general information only and does not constitute legal advice. The consequences of using an AI platform depend on the circumstances, the information involved, the applicable law, and the terms and safeguards of the service.

Written by: Jacob Goldenberg

References/Footnotes:

1. Solosky v The Queen, 1979 CanLII 9 (SCC), [1980] 1 S.C.R. 821, available at: https://canlii.ca/t/1mjtq.
2. Blank v Canada (Minister of Justice), 2006 SCC 39, [2006] 2 S.C.R. 319, available at: https://canlii.ca/t/1p7qn.
3. Suzana Popovic-Montag, “Preserving Solicitor-Client Privilege in the Age of Generative AI: Emerging Lessons from Abroad and Potential Implications for Canadian Law,” The Probater, Vol. 32, No. 2, June 15, 2026, available at: https://hullandhull.com/2026/06/the-probater-vol-32-no-2-june-2026-preserving-solicitor-client-privilege-in-the-age-of-generative-ai-emerging-lessons-from-abroad-and-potential-implications-for-canadian-law/; and David Yi, “Privilege in the Age of Generative AI,” Norton Rose Fulbright, June 2026, available at: https://www.nortonrosefulbright.com/en-ca/knowledge/publications/a2ecd008/privilege-in-the-age-of-generative-ai.
4. United States v Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026), available at: https://www.akingump.com/a/web/ssTGsd5NHbtZ1onzXQMTye/1_25-cr-503-27-memorandum.pdf.