AI Regulation Tracker  /  Ethics guidance issued

Alabama's bar tells lawyers that picking the wrong AI tool is itself a competence failure

Formal Opinion 2026-01 creates no new duties. It maps competence, confidentiality, supervision, candor, fees, and client communication onto generative and agentic AI, and it puts vendor due diligence and documentation inside Rule 1.1. Two Alabama decisions sit in the opening citation.

By Anthony Guerriero  ·  Published July 31, 2026  ·  Last updated July 31, 2026

The Alabama State Bar has become one of the jurisdictions to put agentic AI, not only generative AI, inside a formal ethics opinion. Formal Opinion 2026-01, "Artificial Intelligence Use: Best Practices Under Existing Professional Conduct Rules," is published by the Office of General Counsel and applies the existing Alabama Rules of Professional Conduct rather than proposing new ones.

The opinion states its own theory plainly. Artificial intelligence does not create new ethical duties for lawyers. It recontextualizes longstanding professional obligations, competence, confidentiality, supervision, candor, reasonableness of fees, and client communication, within a new technological environment. It cites ABA Formal Ethics Opinion 512 (2024) for the proposition that lawyers remain fully responsible for the work on behalf of the client regardless of whether AI tools were used in its creation.

The opening citation is local

The opinion opens with a cautionary example rather than a rule. It cites Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), the case in which a lawyer filed a brief citing more than a dozen nonexistent ChatGPT-generated decisions. It then adds two citations closer to home: Ibach v. Stewart, No. SC-2024-0678, 2026 Ala. LEXIS 45 (Ala. Apr. 24, 2026), and Johnson v. Dunn, 792 F. Supp. 3d 1241 (N.D. Ala. 2025).

That pairing is the signal. An Alabama lawyer can no longer treat AI citation failures as a problem that happens in other states. The bar has placed an Alabama Supreme Court decision from April 2026 and a Northern District of Alabama decision from 2025 at the front of its guidance.

Competence now includes choosing the tool

Section III addresses Rule 1.1. The opinion is careful to say the standard does not require technical expertise. Drawing on ABA Opinion 512, it says lawyers need not become AI engineers but must have a reasonable understanding of what an AI tool does, what it does not do, its known limitations and risks, and how its outputs may fail.

Then it extends the duty to procurement. The opinion states that vendor selection is also a competence issue, and that selecting an AI tool without evaluating its accuracy or data-retention policies may itself constitute a failure of competence. A lawyer who deploys an AI tool without basic due diligence on how that tool works and what it can get wrong has not satisfied Rule 1.1.

This is the provision most likely to change firm behaviour. It moves the ethical question upstream of any particular filing. The relevant failure can occur at the moment of purchase, before the tool has produced a single output.

The opinion also treats competence as ongoing rather than static. Because the technology evolves quickly, lawyers must periodically reassess the tools they use, including updates to training data, data-retention practices, and output reliability. Its best-practice list asks lawyers to treat AI tools the way they would evolving e-discovery software or cloud platforms, to fold AI literacy into annual CLE planning, and to review a tool's published accuracy rates, data-retention policies, and known limitations before deploying it.

The documentation ask

One best practice in Section III deserves separate attention because it is the most auditable thing in the opinion. It asks lawyers to document their AI use in client matters, with records including which tool was used, the date, what the tool was asked to do, what output was generated, and what verification steps the supervising lawyer took.

The opinion is guidance, not a rule amendment, so this is not framed as a mandatory record. But a firm that keeps such a record is in a materially different position when a court asks how a fabricated citation reached a filing. The opinion's own glossary makes the point sharper: it defines "discoverable" and notes that AI prompts, drafts, and interaction logs may be discoverable in some circumstances. Firms should decide deliberately what they log and how long they keep it, because the same record that demonstrates diligence is potentially producible.

Verification and the ban on blind reliance

Section IV covers Rules 1.1, 3.1, and 3.3, and states what it calls the clearest ethical consensus nationwide: AI output must never be accepted at face value. It cites ABA Opinion 512 for the position that submitting AI-generated work without independent verification can violate at least Rules 1.1 and 3.3, and identifies three acute contexts, filing pleadings or briefs, citing legal authority, and making representations to courts or third parties.

The standard it sets for review is substantive. An AI draft should be treated no differently than a first draft from a first-year associate. It requires substantive review, not just proofreading, and the reviewing lawyer must bring independent legal judgment rather than simply confirming that the output looks right.

The associated best practices are specific: independently verify every AI-generated legal citation against Westlaw, Lexis, or a comparable authoritative database before filing; run AI-generated factual summaries against the underlying source documents; have a second lawyer or senior paralegal review AI-assisted briefs and pleadings before submission; treat AI drafts as starting points; and never delegate legal judgment, negotiation, or strategic decision-making to AI.

The opinion also spends a short passage on prompt quality, noting that vague or poorly crafted prompts increase the risk of hallucinations and off-point output, and encouraging lawyers to provide context, specify the jurisdiction and applicable rules, and ask the AI to identify its sources.

Why agentic AI gets its own treatment

Section II distinguishes generative AI from agentic AI and explains why the second is the harder ethical problem. Generative AI responds to a prompt and produces content, requiring human follow-up to act on the output. Agentic AI independently plans and executes tasks and takes actions, operating with minimal or no human oversight.

The opinion's illustration is worth repeating because it is the one most partners will recognise. A generative tool drafts a discovery response when prompted. An agentic system might autonomously draft the response when it detects that a new file is opened, email it to the client for review, calendar the filing deadline, and update the case management system, all without further attorney input. The opinion characterises the risk difference accordingly: hallucination in output for generative AI, unsupervised errors with real-world consequences for agentic AI.

That framing matters for the supervision rules. An agentic system that emails a client or calendars a deadline is performing work that, done by a non-lawyer assistant, would sit squarely under Rules 5.1 and 5.3.

Browse the full AI Regulation Tracker →

Drafted with AI assistance and verified against the primary source by a human editor. Informational only, not legal advice.