Client Confidentiality When Lawyers Use AI Tools

Model Rule 1.6 protects information relating to a representation, regardless of source, subject to informed consent, implied authorization, and rule exceptions. ABA Formal Opinion 512 explains that lawyers should make reasonable efforts to prevent inadvertent or unauthorized disclosure or access when using generative AI. Its 2024 analysis says informed consent is required before entering representation-related information into self-learning tools that may disclose that information; it also says consent is not needed for use that does not input such information, such as generic idea generation. The opinion recommends understanding the tool’s terms, privacy policies, retention, access, and how inputs may be used. A generic engagement-letter clause may not amount to informed consent; where consent is required, the client needs adequate information about purpose, risks, alternatives, and potential access. Model Rule 1.4 may separately require communication depending on materiality, client expectations, fee, or the tool’s influence on a significant decision. ABA opinions are guidance under the Model Rules; binding duties depend on rules adopted in the lawyer’s jurisdiction and facts of the matter. Anonymization can reduce risk but may fail when facts, dates, locations, or unusual circumstances identify a client indirectly. Data controls also vary across products and plans, and settings can change. Use approved systems, minimize information, verify retention and training terms, limit access, and preserve human review. Do not assume that a closed workspace automatically makes every prompt safe or privileged.