Client confidentiality and AI tools - a checklist for lawyers

Short answer: ABA Model Rule 1.6(c) requires reasonable efforts to prevent unauthorized access to or disclosure of client information. For AI tools that means doing and recording vendor due diligence, sorting information by sensitivity, limiting what you enter, writing a team policy and planning for a breach. ABA Formal Opinion 512 adds that you need the client's informed consent before putting their confidential information into a tool that may learn from it.

This article is general information, not legal or ethics advice. Check your own state's rules and opinions.

The rules in brief

Several states have their own AI guidance, including the State Bar of California's Practical Guidance for the Use of Generative AI (November 2023) and Florida Bar Ethics Opinion 24-1 (January 2024).

Step 1: Do (and record) vendor due diligence

Opinion 477R expects reasonable due diligence on vendors that handle client information. The detailed questions to ask (training on your data, subprocessors, retention, storage location, two-factor sign-in, encryption, breach notice) are in our guide 12 questions to ask before buying legal research software. What the ethics rules add is keeping a record: save the vendor's answers, the terms version you reviewed and the date, so you can show your efforts were reasonable.

Step 2: Sort information by sensitivity

Comment [18] makes sensitivity the first factor. A simple three-level rule helps staff decide:

Step 3: Get informed consent when it's needed

Opinion 512 says informed consent is needed before entering confidential information into a self-learning tool, and that boilerplate in an engagement letter isn't enough. Consent should explain, in plain words, the tool, what information goes in, the risks (including reuse by the tool) and the benefits. Record it in the file.

Step 4: Write it down for the whole team

• A short AI-use policy: approved tools, what each may be used for, and the information levels allowed (Rules 5.1 and 5.3).
• Training for lawyers and staff when they start and when tools change.
• A rule against personal or free AI accounts for client work.
• Offboarding: remove departing staff from every tool.

Step 5: Plan for something going wrong

ABA Formal Opinion 483 (2018) covers a lawyer's duties after a data breach, including acting promptly to stop it, investigating and telling affected clients. Know who at the firm handles an incident and how you would contact each vendor.

Step 6: Review once a year

Vendors change their terms, add AI features and switch subprocessors. Re-check approved tools at least once a year and whenever a vendor announces material changes.

A low-risk default

For legal research, the question is usually about the law, not the client. "Slip-and-fall in a grocery store, liquid on the floor for an unknown time, Texas" gets the same research results as a version with the client's name. Removing identifiers is the simplest safeguard. It costs nothing and doesn't depend on any vendor's promises.

Frequently asked questions

Can I use ChatGPT or other public AI tools with client information?

It depends on the tool's terms and your safeguards. Opinion 512 asks you to assess the risk of disclosure first. For tools that may learn from inputs, it calls for informed client consent before entering confidential information. Many lawyers avoid entering identifying details at all.

Do I need client consent to use AI for legal research?

Not always. If no confidential information goes into the tool, the confidentiality concern largely falls away. Opinion 512 also discusses when you need to tell clients about AI use under Rule 1.4, for example when it affects fees or the client asks.

Is a vendor's SOC 2 report enough?

It's useful evidence of security controls, but it doesn't answer whether your data is used for training or who can access it. Read the terms and privacy policy too.

Does Rule 1.6(c) require perfect security?

No. It requires reasonable efforts. Comment [18] says unauthorized access or disclosure doesn't by itself mean the lawyer failed to make reasonable efforts.

Sources

• ABA Model Rules of Professional Conduct, Rule 1.6 and Comment [18]
• ABA Formal Opinion 477R, Securing Communication of Protected Client Information (May 11, 2017)
• ABA Formal Opinion 498, Virtual Practice (March 10, 2021)
• ABA Formal Opinion 483, Lawyers' Obligations After an Electronic Data Breach or Cyberattack (October 17, 2018)
• ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024)
• State Bar of California, Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (2023)
• The Florida Bar, Ethics Opinion 24-1 (2024)