The duties don’t change; the tool does
Professional-responsibility guidance on AI has converged on a simple point: using a generative AI tool does not suspend a lawyer’s existing obligations. Competence now includes a working understanding of how an AI tool handles client information. Confidentiality still requires reasonable efforts to prevent unauthorized disclosure. Candor to the court still means you are responsible for what you file, whatever produced the draft.
Does putting client information into an AI tool risk confidentiality or privilege?
It can. Guidance from bar authorities has emphasized that a lawyer must ensure confidential client information placed into an AI tool remains secure and protected from unauthorized access or disclosure. And courts have begun to treat the tool’s terms of service as material: where a consumer AI tool’s terms permit data retention and disclosure to third parties, a court has found that using it can undercut the confidentiality on which attorney-client privilege and work-product protection depend.
This article is general information, not legal or ethics advice. Every lawyer must evaluate these questions under the rules of their own jurisdiction.
Why the terms of service are part of the analysis
The practical consequence is that “is this tool any good?” is not the only question. “What do this tool’s terms let the provider do with what I put in?” is now part of the competence and confidentiality analysis. A tool that retains inputs, uses them for training, or reserves the right to disclose them to third parties is a different risk proposition from one that does not — and from a system where the data never leaves the firm at all.
On-premises architecture changes the math directly. If client data is processed on hardware the firm owns and never transmitted to a third party, an entire category of confidentiality and privilege questions is answered by design rather than by reading a vendor’s fine print.
“The AI said so” is not a defense
The duty of candor and the duty of supervision both land in the same place: the lawyer is responsible for the work. A fabricated citation is the lawyer’s fabricated citation once it is filed. A system that produces confident, unverified output is not a shield; it is an exposure. This is the strongest argument for architecture that verifies claims against source records and fails loudly when it cannot — the safeguard is what lets a lawyer supervise the tool rather than trust it.
The adult-in-the-room posture
None of this is a counsel of despair about AI. It is a counsel of care. The firms that will use these tools well are the ones that treat the professional- responsibility questions as design requirements: confidentiality by architecture, verification by default, a human in the chair for every consequential output. That posture is not a constraint on the technology. It is what makes the technology safe enough for a court file.
This article is general information from a technology consultancy, not legal advice, and does not create an attorney-client relationship. Figures describing the founder’s own practice are illustrative, not a promise of results.