home  /  ai law & rules  /  bar guidance
rules · ai for legal practice

Bar guidance, state by state.

A growing body of guidance that broadly agrees on the duties and disagrees on what they require.

begin here

Where is your firm?

Start a conversation with the AI Adoption Concierge, already scoped to bar guidance. Pick a starting point, or describe your situation directly.

AI Adoption Conciergebar guidance · orientation, not legal or ethics advice
Tell me your jurisdictions and I'll describe what authorities there have published and when. I'll be explicit about what I can't verify. What those rules require of your practice is a question for your own professional-responsibility counsel.

The starting point for most firms is ABA Formal Opinion 512, issued 29 July 2024 — the first ABA ethics guidance on generative AI, and the document most state guidance is written against. It interprets existing Model Rules rather than creating new ones, which is why its conclusions read as familiar duties applied to unfamiliar facts. Since then a substantial number of state and city bars have published opinions, guidance documents, task-force reports or practical guides. They agree on the architecture — competence, confidentiality, supervision, candour, reasonable fees — and diverge on the operative question of what a lawyer must do before putting client information into a tool. Some frame client consent as required; others as recommended; at least one has expressly declined to go as far as the ABA. That divergence is the reason a national summary is misleading.

mechanisms

The duties every authority addresses.

The framework is consistent across jurisdictions. What each requires in practice is not.

Competence

Understanding what the tool does and where it fails — including its propensity to fabricate — and keeping that understanding current as the tools change.

Confidentiality

The central divergence. What must be understood about a tool's retention and training behaviour before client information goes into it, and whether client consent is required or advisable.

Communication

When a client must be told. No authority has imposed a blanket disclosure duty; several identify circumstances that trigger one.

Candour to the tribunal

The obligation that produced the sanctions record. Output reviewed before it is filed, every time.

Supervision

Firm-level obligations — policies, training, and responsibility for what people and vendors do with the tools.

Reasonable fees

Whether time saved may be billed, how tool costs are treated, and whether learning time is chargeable. Authorities are unusually specific here.

methodology

What the evidence shows — and what we examine.

How to read guidance without over-reading it.

Start with your own jurisdictionsAll of them. Firms practising across state lines tend to operate to the strictest, because a single firm-wide standard is easier to follow than a matrix.
Read the instrument, not the summaryA formal opinion, a task-force report and a bar-journal column are different weights of authority and are routinely cited as if identical.
Check what superseded whatAt least one significant guidance document has been replaced by a later edition. Practitioner pages citing the earlier version are still circulating.
Take advice on applicationWhat the rules require of your practice is a question for professional-responsibility counsel, not a research summary.
what's at stake

What turns on getting this right.

Mostly, whether the firm can explain its position when someone asks — a client, an insurer, or a regulator.

a defensible firm standard how client material is handled what clients are told, and when how AI-assisted work is billed answers to client due-diligence questions what the firm requires of its people

Boilerplate consent has been addressed directly.

ABA Formal Opinion 512 states that adding general provisions to an engagement letter purporting to authorise generative AI use is not sufficient to obtain informed consent. Firms that added a clause in 2024 and considered the question closed should read that passage again.

common questions

Bar guidance — practical questions.

Does my client have to consent before I use AI?

It depends on the tool and on your jurisdiction, and authorities have not landed in the same place. ABA Formal Opinion 512 concluded that for self-learning tools — where inputs may train the model or become accessible to others — a client's informed consent is required before inputting information relating to the representation, and that boilerplate engagement-letter language does not achieve it. Some state guidance has framed consent as recommended rather than required; at least one expressly noted the ABA and another state went further than it was prepared to. The practical consequence is that the analysis turns heavily on what the specific tool does with inputs, which is a contract and configuration question before it is an ethics question.

Do I have to tell clients I used AI?

No authority reviewed has imposed a blanket duty. Several identify circumstances that trigger one — where the client asks, where the engagement terms or outside counsel guidelines require it, where consent is needed for confidentiality reasons, where AI use is relevant to the fee, or where output will influence a significant decision. Separately, a number of corporate clients have written disclosure obligations into their outside counsel guidelines, which is a contractual duty rather than an ethical one and is frequently the one that actually binds.

Can I bill for time AI saved me?

Authorities have been unusually direct that a lawyer billing hourly may charge only for time actually spent. The commonly given illustration is that time spent inputting information and reviewing output is billable; time the tool eliminated is not. Guidance has also addressed flat fees — noting it may be unreasonable to charge the same flat fee for work the tool has made substantially faster — and the treatment of tool costs, which several authorities direct toward overhead unless the client has agreed otherwise in advance. This is one of the few areas where the guidance is specific enough to act on directly.

How do I keep track of this?

Assume it changes. New guidance has continued to appear through 2025 and 2026, at least one major guidance document has been replaced by a revised edition addressing agentic AI, and proposed amendments to conduct rules have been circulated for comment in at least one large jurisdiction. A firm operating in several states benefits from a named owner who checks the relevant bars on a schedule rather than relying on the material surfacing. Anything undated is unusable.

related

Related specialization areas & resources.

Mapping guidance across your jurisdictions?

Tell the Institute where you practise. We will help you assemble the picture — your own counsel decides what it requires of you.

AI adoption conciergeorientation · not legal or ethics advice
Tell me your jurisdictions and I'll describe what authorities there have published and when. I'll be explicit about what I can't verify. What those rules require of your practice is a question for your own professional-responsibility counsel.