Before You Paste Client Information Into ChatGPT

Disclosure: Artificial intelligence assisted with drafting and structural refinement. The author reviewed and verified the final content and accepts responsibility for it.

Disclaimer: I am an AI integration strategist, not a lawyer. This article addresses information governance and responsible AI use. It is not legal advice. Requirements vary by jurisdiction, client, matter and forum.

Before You Paste Client Information Into ChatGPT

Would I be comfortable explaining this prompt—and the account that received it—if privilege or confidentiality were later challenged?

Many lawyers experience an AI chat as private and informal, more like working in a blank document than sending information elsewhere.

But a personal AI account should not be assumed to be a private extension of the lawyer’s desk.

When information enters an AI system, it crosses the Prompt Boundary: the point at which it becomes subject to that environment’s contractual, technical and operational conditions.

That boundary is easy to miss. There is no recipient’s name, attachment warning or external email address.

There is only a helpful-looking text box.

AI Makes Disclosure Feel Like Private Thinking

Lawyers are trained to protect confidential information and disclose it only when authorized, necessary and appropriately controlled.

Generative AI can bypass the ordinary cues that reinforce that instinct.

The system often performs better when given more context, encouraging users to explain the complete problem, paste the document or provide the surrounding facts. The same context that improves the response may also increase the sensitivity of the disclosure.

The risk arises because AI can make disclosure feel like private thinking.

Confidentiality Risk Is Broader Than Waiver

Confidentiality, privilege, privacy, cybersecurity and professional competence are related but distinct.

A court may not ultimately find that privilege was waived, yet the disclosure may still create confidentiality, privacy, contractual, security or professional-conduct concerns.

The Law Society of British Columbia cautions that entering privileged information into a generative AI system may put privilege at risk or support a waiver argument.[1] The Canadian Bar Association and Law Society of Alberta also emphasize confidentiality, appropriate system selection, supervision and verification.[2][3][4]

The key question is whether the specific account, configuration, information type and proposed use were assessed and approved.

Removing the Name May Not Be Enough

A prompt can reveal client instructions, legal advice, litigation strategy, settlement posture, investigation findings, regulatory exposure or commercially sensitive facts.

Removing names may not de-identify the matter. Dates, transaction amounts, roles, locations, allegations and unusual facts may still reveal the client, dispute or strategy.

Canadian privacy regulators emphasize accountability, appropriate purposes, data minimization and limiting the disclosure of personal or sensitive information.[5]

Before crossing the Prompt Boundary, consider whether the task can be completed with public information, a fictional scenario, abstracted facts, a generic clause or a smaller extract.

Convenience is not necessity.

Can Clients Put Privilege at Risk by Using?

Clients may create confidentiality or privilege risk by entering legal advice, draft pleadings, settlement discussions or sensitive matter details into an AI system. They may upload a demand letter, ask whether they should settle or paste counsel’s advice into a separate processing environment without appreciating the potential consequences. The legal effect will depend on the facts, the system, the safeguards and applicable law.

As a practical risk-control measure, lawyers should consider advising clients not to upload privileged or confidential matter information into AI systems without first discussing the proposed use.

Four Questions at the Prompt Boundary

1. Has this use been assessed and approved?

Has the firm assessed the specific account, configuration, information category and task?

Approval of a provider does not authorize every feature or use.

2. Does the task require this information?

Can the objective be achieved with less information, abstracted facts or non-client material?

Could the remaining details still identify the client, transaction, dispute or strategy after names are removed?

3. What happens after submission?

Where may the information be retained? Who or what may access it? Can it be used for another purpose? Are memory, sharing or integrations enabled?

The firm should have a defensible basis for concluding that the information is permitted in that environment.

4. How will the result be used and checked?

Who will review the output, and what must be confirmed before reliance?

The level of review should reflect the intended use, the consequence of error and the reliability of the available source material.

Controlled Processing Does Not Establish Accuracy

An AI system can operate under appropriate controls and still produce false, incomplete or misleading output.

The Federal Court requires a declaration where documents submitted to the Court contain content created or generated directly by AI, while distinguishing that from AI-assisted suggestions, recommendations or critiques considered and implemented by a human.[6]

Fluency is a system capability. Accuracy requires human verification.

The Rule to Remember

Do not paste client-identifiable, confidential or privileged information into an AI environment unless the specific account, configuration, information type and use have been assessed and approved.

Before information crosses the Prompt Boundary, the lawyer should be able to explain why the information was necessary, why the environment was appropriate, what would happen after submission and how the output would be checked.

Generative AI can subtly change the lawyer’s instinct from protecting information to volunteering it.

Every AI prompt containing client information is a disclosure decision before it is a productivity decision.

References

[1] Law Society of British Columbia, Guidance on Professional Responsibility and Generative AI.

[2] Canadian Bar Association, Ethics of Artificial Intelligence for the Legal Practitioner.

[3] Law Society of Alberta, The Generative AI Playbook.

[4] Law Society of Alberta, Gen AI Rules of Engagement for Canadian Lawyers.

[5] Office of the Privacy Commissioner of Canada and Canada’s provincial and territorial privacy regulators, Principles for Responsible, Trustworthy and Privacy-Protective Generative AI Technologies.

[6] Federal Court, Notice to the Parties and the Profession: The Use of Artificial Intelligence in Court Proceedings, amended May 7, 2024.

Professional guidance, court directions, product features and vendor terms may change. Confirm the current version before relying on a specific requirement.