Lawyer AI Disclosure to Clients: What State Ethics Guidance Actually Requires

A lawyer does not have to announce every use of generative AI to every client. Disclosure is usually required when the tool receives confidential client information, when AI use is material to how the legal work is performed, when the client asks, or when the lawyer plans to charge the client for AI-generated work in a way the client could reasonably misunderstand. The exact answer changes by state, so a lawyer should start with the ethics guidance that applies where the matter is being handled.

The practical question behind lawyer AI disclosure to clients is not “Did the lawyer use AI?” It is “What information went into it, what did it produce, who can retain that information, and does the client need that information to make an informed decision about representation?” Those are the points covered in understanding AI Ethics Rules for Lawyers.

Clients can check one thing in five minutes without any legal research: look at the engagement letter, fee agreement, or initial email from the lawyer. Search for words such as “technology,” “third-party vendors,” “confidentiality,” “costs,” “legal research,” “automation,” or “artificial intelligence.” If the agreement says the firm may use outside technology providers but says nothing about confidential information being entered into public AI tools, that does not answer every ethics question, but it tells the client what was disclosed at the start.

When Does a Lawyer Need AI Disclosure to Clients Before Using the Tool?

The clearest disclosure issue arises when a lawyer wants to enter client confidences into a generative AI system. A prompt containing a client’s name, medical diagnosis, settlement position, draft contract, witness statement, Social Security number, account number, or a page from an unfiled complaint can reveal protected information even if the prompt does not say “confidential.” A lawyer has to know whether the AI provider stores prompts, uses them to train a model, allows human review, or retains them for 30 days, 90 days, or longer.

The American Bar Association addressed this in Formal Opinion 512, issued in July 2024. The opinion does not impose a blanket rule requiring lawyers to disclose all generative AI use. It says lawyers should obtain informed consent before entering client information into a self-learning generative AI tool when there is a meaningful risk that the information could be disclosed outside the representation.

Informed consent means more than burying one sentence in a 12-page engagement agreement. The lawyer should explain the material facts in plain language: what tool is being used, what client information may be entered, whether the provider can retain or train on that information, what safeguards are in place, and what practical risk remains. If the lawyer can use a closed enterprise system with settings that prevent training and outside access, the consent analysis may be different from using a free public chatbot on a personal account.

Disclosure can also be required even where no confidential facts are entered. If a lawyer relies on AI to prepare a first draft of a demand letter, summarize 800 pages of records, or generate research leads, the client may need to know if that use materially affects the representation, timing, cost, or quality of the work. A lawyer still must review the output line by line; a polished paragraph with a made-up case citation is not a harmless drafting shortcut.

How California, Florida, Pennsylvania, and New York Approach AI Use

California’s State Bar Standing Committee on Professional Responsibility and Conduct addressed generative AI in Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law, published in November 2023, followed by Formal Opinion 2024-1. California does not treat every AI-assisted task as something that must be disclosed. Its guidance focuses on competence, confidentiality, supervision, candor, and billing. A lawyer may need informed consent when AI use creates a significant risk of disclosing client information or when the client needs the information to make an informed decision about the representation.

Florida Bar Ethics Opinion 24-1, issued in January 2024, is more direct about confidential information. It says a lawyer must obtain informed consent before using a generative AI program if confidential information will be entered into the program and the lawyer cannot reasonably determine that the information will remain confidential. Florida’s guidance puts real weight on checking the provider’s terms of service, privacy policy, retention practices, security measures, and training settings before typing facts into the prompt box.

Pennsylvania Bar Association Formal Opinion 2024-200 takes a risk-based approach. It does not say that generative AI use always requires client consent. It says lawyers must make reasonable efforts to prevent unauthorized disclosure of confidential information and should obtain informed consent where the tool creates a risk that confidential material will be revealed. The difference matters: a lawyer using AI to turn a public court opinion into a timeline is in a different position from a lawyer uploading a client’s unredacted psychiatric records.

New York State Bar Association Ethics Opinion 1240, issued in 2024, also rejects a one-size-fits-all disclosure rule. It says lawyers may use generative AI if they protect confidential information, supervise the work, and remain competent. New York specifically identifies situations where disclosure may be needed: when the client asks about AI use, when AI use is material to the representation, or when the lawyer’s charges for AI-assisted work would otherwise be misleading or unreasonable.

What Clients Should Ask Before Agreeing to Generative AI Use

A useful client question is: “Will any of my private documents, messages, medical records, financial records, or case facts be entered into a public or third-party AI system?” That question gets past vague answers about “technology.” A second question is: “Does the provider use prompts or uploaded files to train its system?” A lawyer should be able to answer from the vendor’s actual settings and contract terms, not from an assumption based on a product advertisement.

Ask whether the firm uses an enterprise account or a consumer account. Enterprise tools may offer controls such as no model training, limited retention, encryption, administrator access rules, and a written data-processing agreement. Those features do not automatically solve the confidentiality issue, but they are very different from pasting a client’s facts into a free chatbot whose default setting allows provider retention.

Ask who checks the work before it is sent out. A lawyer should review AI-generated citations, quotations, dates, party names, statutes, calculations, and factual statements. If a draft says a hearing is on March 14, a filing deadline is 21 days away, or a contract allows a 5% late fee, those details need confirmation against the actual court notice, rule, contract, or source document.

Billing deserves its own question: “Will I be charged for lawyer time spent reviewing AI output, for the AI subscription, or both?” A lawyer can charge for legal judgment and necessary review, but the bill should not make 15 minutes of automated first-draft work look like 4.0 hours of original drafting. Clients should receive enough information to understand whether a charge is for a vendor expense, attorney review, document analysis, or a combination of those items.

What Disclosure Should Look Like in an Engagement Letter or Conversation

A workable disclosure does not need to be written like a software contract. It can say that the firm may use approved AI tools for tasks such as organizing documents, creating draft outlines, identifying issues for lawyer review, or summarizing records; that lawyers remain responsible for the final work; and that confidential information will not be entered into systems that use it for public model training without the client’s informed consent. If the firm expects to use a particular tool with sensitive case facts, naming the tool and its data-handling limits is clearer than saying only “advanced technology.”

The disclosure should match the actual job. A family-law matter involving 36 months of bank statements, a criminal matter involving police reports and body-camera footage, and a business matter involving a 60-page asset purchase agreement present different confidentiality concerns. A generic sentence copied into every engagement letter may not be enough when the lawyer expects to upload a large client document set to an outside provider.

The client does not need a technical lecture about neural networks. The client does need a straight answer about what leaves the firm’s system, who may hold it, whether it can be reused by the provider, and what alternative is available if the client does not consent. In a matter involving privileged communications, the related issue is often described as AI chatbot attorney client privilege, especially where a lawyer is considering putting unfiled strategy, interview notes, or settlement analysis into a chatbot prompt.

If a client is uncomfortable with the proposed use, the lawyer should discuss options before the information is entered: use no generative AI for that matter, use the tool only on public or redacted material, use a system with stronger contractual privacy controls, or perform the work through traditional research and drafting. The right choice depends on the matter, the tool, the information, and the ethics guidance that applies to that lawyer.