What is Cowork? Use Cases Blog Services AI Readiness Quiz FAQ Book a Call

Legal · Professional Responsibility

Can US attorneys ethically use AI tools?

ABA Formal Opinion 512, the Model Rules, and state bar guidance, translated into the controls a US law firm actually needs in 2026.

Updated November 2026 · 12 minute read

The short answer

Yes — US attorneys can ethically use generative AI tools, and the organized bar has said so. The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 in July 2024, confirming that generative AI (GAI) is permissible when attorneys discharge their existing duties of competence, confidentiality, communication, candor, fees, and supervision. California, New York, Florida, Texas, New Jersey, DC and others have followed with their own guidance. None require abstention. All require intentionality.

The framework in one line Attorneys do not need a new rulebook for AI. They need to apply the Model Rules they already follow — competence, confidentiality, fees, communication, candor, and supervision — to a new category of nonlawyer assistance.

ABA Formal Opinion 512: the landmark opinion

Issued on 29 July 2024, Opinion 512 is the first comprehensive ABA ethics opinion on generative AI. It does not create new rules. It applies the existing Model Rules to GAI tools and sets out six duties:

  1. Competence (Rule 1.1) — a reasonable understanding of a GAI tool's benefits and risks before use on client matters.
  2. Confidentiality (Rule 1.6) — obtaining informed consent where a tool's handling of client information requires it, and evaluating the tool's terms, security, and training behavior.
  3. Communication (Rule 1.4) — informing clients about material use of AI in their matter where competence or confidentiality require it, and responding candidly to client questions.
  4. Candor to the tribunal (Rule 3.3) — verifying every citation, quotation, and factual representation produced with AI assistance.
  5. Supervisory responsibility (Rules 5.1 and 5.3) — firm-wide policies and training so lawyer and nonlawyer use of AI conforms to the Rules.
  6. Fees (Rule 1.5) — charging only for time actually spent, and disclosing AI-related costs when billed to clients.

Rule 1.1: Competence and the hallucination problem

Comment 8 to Model Rule 1.1 has required attorneys to keep abreast of the benefits and risks associated with relevant technology since 2012. Opinion 512 makes clear this duty now extends to GAI. Attorneys need not become machine learning engineers, but they must understand — at a working level — that large language models can produce confidently stated but fabricated citations, quotations, and legal propositions ("hallucinations").

The sanctions docket is already instructive. Mata v. Avianca (SDNY, 2023), Park v. Kim (2d Cir. 2024), and a growing list of federal district court orders have imposed sanctions on attorneys who filed AI-generated briefs citing nonexistent cases. The competence failure is not using AI. It is filing without verification.

Opinion 512 is equally explicit about confidentiality competence: attorneys must understand how a given tool handles inputs, whether prompts are retained, whether they are used for model training, and where data is stored. A tool that trains on inputs is not categorically prohibited, but it changes the Rule 1.6 analysis.

Rule 1.6: Confidentiality, prompts, and consent

Model Rule 1.6(a) prohibits revealing information relating to the representation of a client without informed consent, impliedly authorized disclosure, or another exception. Pasting client-identifying facts into a GAI tool is a disclosure. The question is whether it is authorized.

Opinion 512 distinguishes between two categories:

Many state bars (NY, FL, CA among them) take a more conservative line: when in doubt, obtain informed consent and document it. The practical answer for most firms is to update the engagement letter to describe AI use in general terms, and to obtain express written consent before using any tool that trains on inputs.

Rule 1.5: Fees, billable hours, and AI-produced work

Opinion 512 is unambiguous on billing: attorneys may not bill clients for time they did not actually spend. If a task that would have taken an hour of associate time is completed with AI in ten minutes of verified, reviewed output, the client is billed for the ten minutes. Billing a phantom hour is a Rule 1.5 reasonableness violation and, potentially, Rule 8.4(c) dishonesty.

Three practical implications:

Rules 5.1 and 5.3: Supervising lawyers, associates, and nonlawyers

Opinion 512 treats a GAI tool as a form of nonlawyer assistance. That triggers Rule 5.3's supervisory framework: partners and managers must make reasonable efforts to ensure the firm has measures in place giving reasonable assurance that AI use conforms to the Rules. Rule 5.1 applies the same duty to supervising associates and junior attorneys.

In practice that means (i) a written AI acceptable use policy, (ii) an approved tool list, (iii) training before use on client matters, (iv) workflows that build in human review of every output, and (v) a clear channel for escalating novel uses. "I trusted my associate's judgment" is not a defense if the firm never articulated what that judgment was supposed to look like.

State variations: CA, NY, FL, TX

Five-step adoption checklist for US firms

  1. Choose an enterprise-grade tool that contractually prohibits training on inputs, offers a DPA, and provides appropriate security. Review the terms with your general counsel or outside ethics counsel.
  2. Publish a written AI acceptable use policy covering approved tools, prohibited uses, prompt hygiene, confidentiality handling, verification standards, and escalation.
  3. Update your engagement letter to describe AI use in your practice and, where your jurisdiction expects it, obtain express informed consent.
  4. Train every user — attorneys, paralegals, legal assistants — on hallucination risk, confidentiality, and verification. Document the training.
  5. Build human review into every workflow. No AI-produced citation, quotation, or legal conclusion leaves the firm without attorney verification. Log the review.

Where this leaves US law firms

Opinion 512 is not a brake. It is a license, conditional on competence and discipline. Firms that build a defensible framework around a vetted toolset will capture meaningful efficiency on attendance notes, client correspondence, contract summaries, and research briefings — the workflows we cover in our US law firm implementation program. Firms that ignore the duties will eventually explain themselves to a disciplinary board, a sanctions hearing, or a malpractice carrier.

FAQ

AI ethics questions US firms ask us

Do US attorneys need client consent to use AI tools?

ABA Formal Opinion 512 (July 2024) concludes that whether informed consent is required depends on the tool and how it handles client information. If a self-learning GAI tool ingests client-confidential information in a way that could disclose it to third parties, express informed consent under Model Rule 1.6(a) is generally required. For tools that do not train on inputs and operate under appropriate confidentiality terms, many jurisdictions permit reliance on implied authorization, though prudent firms obtain informed consent as a matter of practice and document the discussion in the engagement letter.

Can attorneys bill clients for time saved by AI?

Under Model Rule 1.5, fees must be reasonable. ABA Opinion 512 states attorneys may not bill for hours not actually worked. If AI completes a task in ten minutes that previously took an hour, the attorney bills only the actual time spent, including review and revision. Flat fee arrangements may be appropriate where efficiency gains should be shared. Charging clients for AI software costs as a disbursement generally requires disclosure and agreement, consistent with Opinion 93-379.

Which states have issued AI ethics guidance?

California (State Bar Practical Guidance, November 2023), New York (NYSBA Task Force Report, April 2024), Florida (Advisory Opinion 24-1), Texas (State Bar Opinion on generative AI, 2024), and New Jersey, DC, Virginia and Michigan have published guidance. Positions vary on consent, supervision, and billing, but all converge on the attorney's non-delegable duty of competence and confidentiality.

Does using AI create a duty to disclose to the court?

Opinion 512 does not impose a general duty to disclose AI use to courts. However, many federal judges have standing orders requiring certification that AI was not used, or that any AI-generated content was verified. Attorneys must check local rules and judge standing orders before filing. Rule 3.3 candor duties apply to every citation and representation, so AI-produced citations must be independently verified.

What is the biggest AI ethics risk for US law firms in 2026?

Hallucinated citations remain the most publicized risk, producing sanctions in multiple federal districts. The quieter but larger systemic risk is confidentiality: paralegals and associates pasting client-identifying information into consumer AI tools without firm-approved terms, violating Rule 1.6. A written AI acceptable use policy, firm-provisioned accounts, and training on prompt hygiene are the standard controls.

Do Rules 5.1 and 5.3 require supervising AI the same way as associates?

Opinion 512 treats generative AI as a nonlawyer assistant for supervision purposes. Managing partners and supervising lawyers must make reasonable efforts to ensure the firm has measures in place giving reasonable assurance that AI use conforms to the Rules of Professional Conduct — written policies, training, approved tools, and review workflows, not merely trusting individual judgment.

Build a defensible AI framework for your firm

Claudable trains US law firms to adopt AI inside the Model Rules — policy, workflow, verification, and team enablement in a single engagement.

See Training Options Book a Discovery Call