Legal AI and Professional Ethics: What the Duty of Competence and Supervision Actually Require

If you searched for "AI legal ethics" or "ABA Model Rules AI," the underlying question is: what does professional responsibility actually require when I use AI in my practice? The rules were not written with generative AI in mind, but bar authorities have made clear that existing duties — competence, supervision, confidentiality, and candor — apply in full, and courts have already sanctioned lawyers who treated AI output as if those duties did not apply.

The duty of technological competence

Comment 8 to ABA Model Rule 1.1 requires a lawyer to keep abreast of the benefits and risks of relevant technology. As legal AI tools become part of ordinary practice, that duty increasingly means understanding, at a working level, how a given tool can fail — specifically, that generative models can produce fluent, confident, and entirely fabricated citations. A lawyer does not need to understand the underlying architecture, but not knowing that hallucination is a real risk is itself a competence gap.

The duty to supervise extends to AI output

Model Rule 5.3, governing supervision of nonlawyer assistance, has been read by several bar ethics opinions to extend to AI tools used in a lawyer's practice. The lawyer supervising the tool's use bears responsibility for its output in essentially the same way as for work delegated to a paralegal or associate — the tool does not carry independent accountability, and "the AI made the error" is not a defense that has held up in the sanctions cases decided so far.

Confidentiality obligations do not pause for AI

• Model Rule 1.6 requires reasonable efforts to prevent unauthorized disclosure of client information, and that obligation applies to what a lawyer submits to an AI tool, not just what appears in a final work product.
• Know where the data goes: before entering client facts into any tool, confirm whether it retains, trains on, or otherwise processes that input beyond the immediate session.
• Vendor terms matter: a tool's data-handling policy is part of the diligence a lawyer owes the client, not a technical detail to skip.

Candor to the tribunal and the citation-verification duty

Model Rule 3.3 requires candor to the court, and the string of sanctions orders against lawyers who filed AI-generated briefs with fabricated citations shows how directly that duty is now tested by AI use. Verifying every citation an AI tool produces against the actual source is not best practice layered on top of the ethics rules — for a filed document, it is what the existing duty of candor requires in an AI-assisted workflow.

Disclosure to clients: an emerging expectation

Several jurisdictions and firms are moving toward disclosing AI use to clients, particularly where it affects billing (charging full research time for AI-assisted work) or where a client has a reasonable expectation about who is doing the work. Even where no rule explicitly mandates disclosure yet, transparency about AI use tends to track well against the broader duty of communication under Model Rule 1.4.

The takeaway

None of the core duties — competence, supervision, confidentiality, candor — are suspended by the fact that a tool did the drafting. AI can speed up research and document work considerably, but the lawyer using it remains fully accountable for verifying its output, protecting client information, and disclosing its use where a client would reasonably expect to know. Treat AI adoption as a supervision problem to design for, not a shortcut around the rules that already govern the practice.