Artificial intelligence has become a part of daily legal practice. The New Hampshire Bar Association Ethics Committee has published three Ethics Corner articles on this topic. Those articles, in conjunction with ABA Formal Opinion 512, convey a clear message: there is no AI exception to the Rules of Professional Conduct. AI is a technology tool. Lawyers remain responsible for every output we generate using AI. This article addresses four core ethical duties implicated by AI.
Confidentiality and Security. Client confidentiality is a threshold concern. Rule 1.6 bars revealing information relating to a representation without informed consent. The Committee applied that rule in Ethics of Using Artificial Intelligence in Practice (May 15, 2024) and Ethics of Drafting Documents with Artificial Intelligence (Sept. 10, 2024). Both articles warned against entering client information into then-existing AI tools, which might have retained or trained on those inputs.
Those articles reflected the technology of late 2024. They treated “outside” AI as consumer tools. The September article observed that enterprise AI was then “still in its infancy.” It predicted that firms would soon keep AI “in-house” to protect client information.
That future has arrived. Enterprise tools now offer no-training, zero-retention, and deletion terms by contract. The ABA Task Force’s Year 2 Report (Dec. 15, 2025) confirms the shift. Firms are moving from consumer tools, which “present privacy and confidentiality risks,” to secure enterprise tools. Enterprise AI is now the baseline standard of care.
Security controls also are mandatory. Formal Opinion 512 (July 29, 2024) directs lawyers to read each tool’s terms of use and privacy policy, understand how the tool handles data, and consult IT or cybersecurity experts when needed. Just like any other technology, lawyers must ensure that the client information we entrust to AI vendors is protected by appropriate security controls.
The client consent requirement has not evaporated entirely. Rule 1.4 and Formal Opinion 512 make consent fact-specific. For example, consent may be required when AI affects the fee, risks client information, or shapes a significant decision.
Supervision and Candor. Verification is another theme of the Ethics Corner articles. Rule 2.1 requires independent professional judgment. Rules 3.1, 3.3, and 8.4(c) forbid frivolous contentions, false statements to courts, and misrepresentation. Formal Opinion 512 requires lawyers to review all AI inputs and output, including every citation, before filing.
General AI applications, like Claude and ChatGPT, do not access legal databases, like Westlaw and Lexis. They draw only on public Internet sources. As a result, they can fabricate authority. Lawyers therefore cannot rely on generative AI alone to conduct legal research. Courts have sanctioned, and will continue to sanction, lawyers and litigants for filing fictitious citations hallucinated by AI. See Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023); Park v. Kim, 91 F.4th 610 (2d Cir. 2024); Kohls v. Ellison, 2025 WL 66514 (D. Minn. Jan. 10, 2025); Morgan v. Community Against Violence, 2024 WL 218410 (D.N.M. Jan. 19, 2024).
However, used correctly, AI applications are highly effective research aids. Lawyers conduct the research and load the authorities into the tool. The tool reads, summarizes, analyzes, and issue-spots across that dataset. Directed this way, the tool is not performing the research. The danger of hallucination is avoided. AI is also being integrated into the leading research databases, including Westlaw and Lexis. That makes research with those databases far more efficient. While the lawyer must still read the cited authorities and verify each quote and citation before any external use, the usefulness of the technology as a research and writing aid is powerful and undeniable.
The October 2024 Ethics Corner article makes a related point. It grounds bias awareness in competence and independent judgment. It also cautions against automation bias.
Rules 5.1 and 5.3 require lawyers to set AI policies and to train and oversee users of AI. And Rule 5.3 now reaches nonlawyer “assistance,” human or not. Supervise an AI tool as you would a paralegal or junior lawyer. Always keep a human in the loop. And implement a written AI acceptable-use policy now, before a client or a court asks what yours is.
Billing. Rule 1.5 requires reasonable fees, and AI tests that duty. The September 2024 Ethics Corner article noted that failing to use cost-reducing technology could render a fee unreasonable. Formal Opinion 512 sets the limits. A lawyer billing hourly must bill actual time. A lawyer may not bill for time the tool saved. A flat fee may be unreasonable if AI makes the work materially faster. And a lawyer may not charge a client to learn a tool the lawyer uses regularly.
The opinion also separates overhead from expense. A grammar checker built into word processing is overhead. A third-party, per-use service may be billed as an expense, as long as that is disclosed and agreed in advance.
Privilege and Work Product. New Hampshire’s Ethics Committee has not yet addressed privilege. However, courts are starting to do so. Protection typically turns on three variables: who directed the use of AI; whether the tool was consumer or enterprise; and whether the claim is privilege or work product.
Counsel-directed use tends to be protected. In Tremblay v. OpenAI, Inc., 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024), counsel’s prompts reflected mental impressions and were opinion work product. In Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), a litigant’s AI use did not waive work product because the model was “merely a tool, not a person.”
By contrast, independent client use tends to forfeit protection. In United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), a defendant’s use of a public AI platform was neither privileged nor work product, because the tool “is not an attorney.”
Courts and parties are writing issues into protective orders. Some require disclosure of the AI tool used. See Morgan v. V2X, Inc., 2026 WL 864223 (D. Colo. Mar. 30, 2026). Others bar uploading discovery into open or public AI tools. See Jeffries v. Harcros Chemicals, Inc., 2026 WL 820218 (D. Kan. Mar. 25, 2026). Counsel clients about these issues at the start of a representation. Address AI use in engagement letters. And fold AI prompts, outputs, and logs into litigation-hold and preservation protocols.
Conclusion. AI may be new, but our ethical duties are not. Lawyers should adopt enterprise-grade tools, verify inputs and outputs, supervise AI use, bill honestly, and preserve privilege and work-product protection.