Can Lawyers Use AI Without Breaking Confidentiality? The 5 Rules That Matter

The 2026 Confidentiality Landscape

The rules aren't one document; they're a fast-growing patchwork. The ABA's Formal Opinion 512 (July 2024) applies Model Rule 1.6 to generative AI. Florida's Opinion 24-1 and Texas's AI Ethics Opinion 705 say the same in state terms: lawyers must protect client confidentiality when using AI. Justia's 50-state survey concludes lawyers should avoid inputting confidential client information into generative AI platforms — including client identifying data. As of 2026, 11 states plus D.C. have formal AI opinions: they agree on competence, verification, and billing, and split on when client disclosure is required.

And the courts have joined in. In United States v. Heppner (S.D.N.Y., Feb 2026), a judge held that documents a defendant generated with a consumer AI chatbot weren't confidential — the platform's terms allowed data collection and disclosure. The lesson isn't confined to criminal cases: whatever you put into a public AI tool is data a court can treat as disclosed.

The Five Rules

1. Vet the tool before any client data touches it. Read the provider's data, retention, training, and disclosure terms. If you can't verify how data is handled — or the terms allow training on inputs — it's not approved for client work. This is the "reasonable understanding" ABA 512 requires under Rule 1.1.
2. Keep client data off public AI. No full contracts, no client names, no identifying data in general-purpose chatbots. The 50-state guidance is unambiguous. If a tool can't guarantee confidentiality, the client's information can't go in it.
3. Prefer local or on-premise AI. When the model runs on your hardware, there is no third party, no training pipeline, and no disclosure clause. The confidentiality duty is satisfied by the architecture instead of by a terms-of-service promise.
4. Get informed consent where required. Rule 1.6 contemplates client informed consent before disclosure of confidential information. Some jurisdictions require or recommend telling clients about AI use. When in doubt, ask — and document.
5. Write the policy and follow it. Under Rule 5.1, firm leadership must establish policies on AI use. An approved-tools list, prohibited-tools list, and verification workflow turn these rules from intentions into practice — and into a defense if a question ever comes up.

A 60-Second Self-Check Before Every Use

Frequently Asked Questions (FAQ)

Can lawyers use AI without breaking confidentiality?

Yes, if the tool and workflow protect client information. ABA Formal Opinion 512 and state opinions (Florida 24-1, Texas 705) all require reasonable efforts to safeguard confidential information — vet the tool's data practices, avoid inputting client data into public tools, get consent where required, and verify output.

What does Rule 1.6 require for AI use?

Rule 1.6 requires lawyers to make reasonable efforts to prevent disclosure of information relating to the representation. Using a tool that collects, trains on, or may disclose inputs without client informed consent can violate that duty.

How many states have AI ethics opinions?

As of 2026, 11 states plus D.C. have formal ethics opinions on lawyers using AI. They broadly agree on competence, verification, and billing, and split on when client disclosure is required — so check your own jurisdiction.

Is it safe to paste client documents into ChatGPT?

No. The 50-state survey by Justia and multiple state opinions say lawyers should avoid inputting confidential client information into generative AI platforms. Use local AI on your own hardware instead, or get informed client consent and verify the provider's data terms.

What is the safest AI setup for a law firm?

Local or on-premise AI: documents indexed and answered on firm hardware, nothing sent to a third party. This satisfies the confidentiality duty by architecture. You still need a firm policy, output verification, and — where required — client consent.