Disclosing AI Use to Clients: What Professional Ethics Codes Actually Require in 2026

"Do I have to tell my client I used AI?" is one of the most common questions professionals ask in 2026, and the answer is genuinely different by profession. There is no general legal obligation to disclose AI use to a client across all professions. There are specific obligations under specific professional ethics codes, state licensing rules, and certain new state statutes — and the obligations have moved meaningfully in 2025-2026.

This guide walks through what each major professional ethics framework requires (or doesn't), where state law adds an obligation on top, and what disclosure language tends to hold up under board review. It is starting orientation, not legal advice; consult your bar / board / compliance counsel for your specific situation.

The general framework most professionals are operating under

In most regulated professions, the obligation isn't a flat "disclose AI use." It's a combination of three duties that AI use can implicate:

1. Competence — the professional remains responsible for the work product regardless of tools used. AI as a tool doesn't reduce the standard.
2. Confidentiality — using AI tools that train on or retain inputs can be a confidentiality breach if client information is sent to a third-party processor not bound by the same confidentiality obligation.
3. Communication — the client is generally entitled to material information about how the matter is being handled, including (in some contexts) AI tools that materially affect the work.

Disclosure obligations typically arise when one of these three duties cannot otherwise be satisfied without the client knowing.

Lawyers — ABA Formal Opinion 512

The American Bar Association's Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on July 29, 2024, addressing generative AI ("GAI") use in legal practice. ABA opinions are persuasive but not binding; state bars may interpret differently.

What Op 512 actually says on client communication (Model Rule 1.4):

• There is no across-the-board duty to disclose GAI use to every client on every matter.
• If the client asks, the lawyer must disclose. ("Disclosure is a must if a client asks.")
• Disclosure is required when GAI use is "relevant to the basis or reasonableness of the lawyer's fee" — this language from Op 512 has been read to require disclosure when the lawyer is using GAI in a way that affects how the client is being billed (e.g., billing for time spent on a task substantially produced by AI). The reasonableness of fees obligation under Model Rule 1.5 connects to this.
• Informed consent must be obtained in advance when client information is being inputted into a GAI tool, "when the use of a GAI tool is relevant to the basis or reasonableness of a lawyer's fee" — and, more broadly, when client confidentiality could be implicated.
• Beyond those triggers, whether to disclose is a factually-specific inquiry. Lawyers must use professional judgment based on the circumstances of the matter.

Op 512 also addresses competence (Rule 1.1 — lawyers must understand the benefits and risks of the GAI tools they use), confidentiality (Rule 1.6 — pre-input risk assessment of disclosure to third parties), candor to tribunals (Rule 3.3 — accuracy and verification of AI-generated content, including citation checking), supervision (Rules 5.1 and 5.3 — applies to AI tools used by the firm), and fees.

State bars matter too: Texas, Illinois, California, and others issued state-level opinions on AI use ahead of Op 512, and Op 512 sits alongside (not above) state opinions. Check your state bar's current opinion.

Physicians, nurses, and other healthcare clinicians

Most state medical boards and the AMA's Code of Medical Ethics do not yet impose a flat AI-use disclosure obligation. The general framework: the clinician remains responsible for clinical decisions; AI is a tool. However:

• California AB 489 (effective January 1, 2026) prohibits AI from using terms or credentials that imply licensure ("doctor," "MD," "RN," etc.) without licensed-clinician oversight, and prohibits marketing language suggesting clinical expertise unless the product is genuinely supported by licensed professionals. It also relates to California AB 3030 (passed 2024), which requires healthcare facilities to disclose to patients when AI-generated correspondence is sent without human review.
• AB 3030's disclosure requirements vary by communication type: prominently at the beginning for written; throughout for chat/continuous online interactions; verbally at start and end for audio; throughout for video. Plus clear instructions on how the patient can reach a human.
• Utah HB 452 (effective May 7, 2025) imposes specific requirements on "mental health chatbots" — AI tools designed to engage in confidential-conversation-style interactions similar to those with a licensed therapist. Restrictions on advertising, requirements on disclosure, and protections for personal information.
• Utah's broader AI Policy Act (AIPA), as amended by SB 226 and SB 332 (effective May 7, 2025), requires disclosure that a consumer is interacting with generative AI rather than a human when asked in a "clear and unambiguous request," and proactive disclosure at the start of any "high-risk artificial intelligence interaction" in regulated occupations — including the provision of medical or mental health advice.

For most clinical workflows where AI is a documentation assistant and a licensed clinician reviews the output before it reaches the patient or chart, the operational risk is low. For workflows where AI-generated communication reaches the patient without human review, AB 3030-style disclosure is the emerging baseline in California and a defensible practice elsewhere.

CPAs and accounting professionals

The AICPA Code of Professional Conduct does not contain an AI-specific disclosure rule as of mid-2026. The framework operates through:

• Integrity and Objectivity (1.100) — the CPA remains responsible for the work product.
• Due Care / Competence (1.300) — the CPA must understand the tools used, including their limitations.
• Confidential Client Information (1.700) — the CPA cannot disclose client information without consent; this implicates how client data is used in AI tools, particularly those that train on or retain inputs.

The practical guidance most AICPA-affiliated firms are operating under: AI use does not require disclosure to the client unless the AI use affects work product reliability, billing, or confidentiality. Firms typically address AI use in the engagement letter rather than per-engagement disclosure.

Investment advisers (SEC-registered) and financial advisors

The SEC has not adopted AI-specific regulations for investment advisers as of mid-2026, but applies the Investment Advisers Act of 1940 to AI use. Key implications:

• Form ADV Part 2A requires disclosure of methods of analysis and investment strategies. If AI materially shapes the formulation of investment advice, the brochure should describe AI-based methods.
• Rule 206(4)-1 (the Marketing Rule) prohibits untrue or unsubstantiated statements. The SEC has charged advisers for "AI washing" — overstating AI capabilities in marketing. Penalties to date include $225,000 (Delphia) and $175,000 (Global Predictions) in 2024 enforcement.
• Rule 206(4)-7 requires written compliance policies tailored to the adviser's business, which now should address AI tools.
• The SEC's 2025 examination priorities flagged AI-related policies and disclosures as a focus area.

FINRA Regulatory Notice 24-09 clarifies that supervision, recordkeeping, and communication standards under FINRA rules apply to AI tools used by broker-dealers. FINRA has not added AI-specific disclosure rules but expects firms to address AI in their supervisory programs.

Practical disclosure for RIAs: address AI use in Form ADV Part 2A where material, address it in client engagement / marketing materials accurately (no AI washing), and address it in the compliance program. Per-transaction disclosure to the client is generally not required.

Mental health therapists (LCSW, LMFT, LPC, psychologist)

State licensing boards govern. Most state mental-health licensing boards have not yet adopted AI-specific disclosure rules, though several are in process. The frameworks that apply today:

• Confidentiality codes (each profession's ethics code — APA, NASW, AAMFT, ACA) apply to any third-party disclosure of client information. Sending client information to an AI tool without a BAA-equivalent and without the client's informed consent is the practical issue.
• Utah HB 452 specifically covers "mental health chatbots" (see above).
• California AB 489 and AB 3030 affect AI-generated patient communications in California.
• Some state boards (e.g., New York, California) have issued informal guidance recommending disclosure when AI tools materially affect care delivery.

Practical baseline: therapists should not enter client information into general-purpose AI tools (most consumer tiers don't offer BAAs and aren't HIPAA-covered). For BAA-covered, HIPAA-compliant workflows where AI is used as a documentation assistant under clinician review, the operational risk is low; disclosure in the practice's notice of privacy practices is a reasonable baseline.

Disclosure language that holds up

When you do disclose, three elements tend to satisfy ethics reviewers and bar/board investigators:

1. What AI was used for. Be specific. "We use Claude (Anthropic) to draft initial responses to discovery requests, which are then reviewed and edited by the attorney handling your matter."
2. What confidentiality protections are in place. "Our use is under a Business Associate Agreement / data processing agreement that prohibits the vendor from training on or retaining your information beyond the immediate request."
3. What human review applies. "All AI-assisted work product is reviewed by [the responsible professional] before it is used in your matter / reaches you / is filed on your behalf."

Avoid disclosure language that:

• Overstates AI capability ("our AI determines optimal strategy") — this is the AI-washing failure mode.
• Understates human involvement ("the AI handled your matter").
• Disclaims responsibility ("the AI made errors, not us"). The professional remains responsible.

What's coming next

State laws are the leading edge. Through 2026, expect:

• More state-level AI disclosure laws (see our state-by-state AI disclosure guide) that add proactive disclosure obligations on top of existing professional codes.
• State bar ethics opinions interpreting ABA Op 512 — different states will weigh the discretionary disclosure question differently.
• State medical and nursing board opinions on AI documentation tools and patient communication.
• Continued SEC enforcement under existing rules; possibly AI-specific SEC rulemaking on disclosure in adviser brochures.

What this guide is — and what it isn't

This is general orientation as of May 2026 for working professionals. It is not legal advice and does not substitute for your state bar / licensing board / regulator's current requirements for your specific situation. Consult your professional responsibility counsel, your state bar / board / regulator's current opinions, and your firm's compliance program.

Related reading: HIPAA-compliant AI guide, AI BAA vendor guide, PHI vs PII, and the state-by-state AI disclosure rules guide.

---

This article is general professional-ethics orientation as of May 2026. ABA Formal Opinion 512 is persuasive, not binding; state bars may interpret differently. State laws referenced (California AB 489 and AB 3030, Utah AIPA, Utah HB 452) are summarized; verify current effective dates and language. SEC and FINRA positions evolve. This article does not constitute legal or compliance advice. Sources: ABA Formal Op 512 (July 29, 2024); California Health & Safety Code § 1339.75 (AB 3030) and AB 489; Utah HB 452; Utah AIPA as amended by SB 226 and SB 332; SEC Marketing Rule 206(4)-1; FINRA Regulatory Notice 24-09; AICPA Code of Professional Conduct.