How to respond to outside counsel guidelines that restrict AI

What do outside counsel guidelines say about AI?

An AI clause in outside counsel guidelines can ban generative AI on the client's matters outright, allow it only with the client's prior written approval, require disclosure of any AI use, limit the data that can go into a tool, or restrict how AI-assisted time is billed. The terms vary by client, so the only safe answer is the one written in each client's own guidelines. Some guidelines combine several of these, and many sit inside a broader security section that also covers encryption, breach notice and vendor approval.

That variety is the operational problem. A 40-attorney firm with a dozen institutional clients can end up with a dozen different AI rules, while its attorneys use the same Microsoft 365 tenant, the same research platforms and the same document management system for all of them.

Do outside counsel guidelines add to a law firm's confidentiality duties?

Yes, they can. ABA Model Rule 1.6(c) requires reasonable efforts to prevent unauthorized disclosure of client information, and Comment [18] says a client may require the lawyer to implement special security measures beyond what the rule requires. Once the firm accepts a guideline, its AI terms sit on top of that baseline.

ABA Formal Opinion 512 is direct on disclosure: lawyers must disclose their generative AI practices if a client asks how they conducted their work, or if the client expressly requires disclosure under the engagement agreement or the client's outside counsel guidelines. The opinion also says informed consent is required before inputting information relating to a representation into a self-learning tool, and that boilerplate language in an engagement letter is not enough. California attorneys carry their own statutory duty under Business and Professions Code section 6068(e)(1) to maintain client confidences inviolate.

How do you find every AI term in your clients' outside counsel guidelines?

Find every AI term by pulling all current outside counsel guidelines, engagement letters and client security addenda into one place and reading each one for AI language. The terms are usually scattered, so include clauses that never use the word "AI" but restrict "third-party tools," "cloud services" or "automated review."

• Record each client's rule in one matrix: prohibited, approval required, disclosure required, data limits, billing limits.
• Note who at the client approves exceptions and how (email, portal, signed form).
• Note the effective date and the next version the client has announced, since guidelines change on renewal.
• Tie each row to the client and matter numbers used in your practice management and document systems.

If you already maintain answers to client security questionnaires, the same owner should keep this matrix. Our guide on answering outside counsel security questionnaires covers that process.

Which AI tools does a client restriction actually reach?

A client restriction on "generative AI" can reach far more than a chatbot: Microsoft 365 Copilot, AI features inside legal research platforms, e-discovery review tools, transcription and meeting notetakers, and AI features added to the document management system. Build a list of every tool and feature in use, then classify each one by the terms it runs under. Our guide to finding AI tools attorneys signed up for with firm email covers the tools IT never approved.

Opinion 512 tells lawyers to read and understand a tool's terms of use, privacy policy and related contract terms. The State Bar of California's practical guidance adds that a lawyer must not input confidential client information into a generative AI product that lacks adequate confidentiality and security protections, and should consult IT or cybersecurity professionals on retention and security. That review tells you which tools a client might accept and which are off the table regardless of what the client says.

When should a law firm ask a client for an AI exception?

A firm should ask for an exception before it accepts the guidelines, whenever a restriction would block work the client is paying for. A blanket ban is often written with public consumer tools in mind. A client may approve a specific enterprise tool if the firm can show what the vendor contract says about training on inputs, retention, access and location of data.

• Ask in writing, name the specific tool and the specific matters, and describe the safeguards.
• Get the approval in writing from the person the guideline names, and file it with the matter.
• Do not treat a general engagement letter clause as consent. Opinion 512 says boilerplate is not adequate.
• Set a review date, since approvals tied to a vendor contract can lapse when the contract changes.

How do you enforce a client-specific AI restriction?

Enforce a client-specific AI restriction with matter-level technical controls plus supervision, because a written rule alone does not stop a busy associate. California Rules of Professional Conduct 5.1 and 5.3 require managerial lawyers to make reasonable efforts to ensure the firm has measures giving reasonable assurance that lawyers and nonlawyers comply.

• Label restricted matters. A sensitivity label applied to the matter's workspace and files gives every other control something to key on.
• Exclude labeled content from Copilot. A Purview DLP policy using the Copilot location can keep labeled files and emails out of Copilot processing.
• Block unsanctioned AI sites on firm devices. Apps marked unsanctioned in Defender for Cloud Apps can be blocked through Defender for Endpoint.
• Keep a record. Retention policies for Copilot and AI apps can capture prompts and responses so the firm can answer a client's disclosure request with evidence.

California adds a deadline. SB 574, signed September 30, 2026 and operative January 1, 2027, adds Business and Professions Code section 6068.1. It bars entering confidential, personally identifying or other nonpublic information into a generative AI system unless access is restricted to the attorney and people the attorney authorizes who are obligated to keep it confidential. The State Bar has also proposed AI-specific comments to rules 1.1, 1.4, 1.6, 3.3, 5.1 and 5.3, which are not in effect until approved.

What should a law firm do about client AI restrictions this quarter?

1. Pull every active set of outside counsel guidelines and extract the AI terms into one matrix.
2. Inventory every AI tool and AI feature in use, including features inside platforms you already license.
3. Compare the two and list each matter where a tool in use conflicts with a client term.
4. For each conflict, either stop the use or request a written exception naming the tool.
5. Label restricted matters and turn on the matching DLP, app blocking and retention controls.
6. Brief attorneys and staff on which clients restrict AI and how to check before using a tool, and fold the client rules into your law firm AI policy.
7. Prepare for SB 574 on January 1, 2027 by confirming which approved tools restrict access to the firm.

How can s90 help a law firm meet client AI terms?

s90 works with law firms across Los Angeles and Orange County to turn client AI terms into controls the firm can show. We inventory the AI tools and features in your environment, map them against your client matrix, and configure labels, DLP, app blocking and prompt retention so a client's restriction applies to that client's matters. When a client asks how AI was used on its work, the firm has the record ready. If that would help, start with a conversation about your current client guidelines.