Start With a Policy, Because Silence Is Already One
A firm that has never discussed AI has not adopted a "no AI" policy; it has simply left AI use unsupervised. Associates draft with a chatbot, paralegals summarize deposition transcripts in a free tool, and nobody tells you because nobody asked. A flat ban does not stop this. It pushes the behavior underground, beyond your ability to see, supervise, or correct it when it goes wrong.
The better approach is to bring AI use into the open before you first learn of it from a sanctions motion.
Hold the Meeting. Ask the Three Questions.
Sit down with everyone in the firm, lawyers and staff alike, and run an honest inventory. Ask three questions: What AI tools are you using? Which is your favorite, and why? What do you use it for?
That inventory is your starting point. It tells you which tasks your team actually wants help with, whether that is first drafts, summarizing records, organizing exhibits, or research, and it tells you which tools are already in the building. From there you can evaluate options deliberately instead of reacting after the fact. Instead of starting from a blank page, you are formalizing what is already happening.
A written policy does not need to be long. It needs to say which tools are approved, what kinds of information may and may not go into them, who is responsible for verifying output, and who to ask when something is unclear. Courts have noticed the difference. Where firms had real AI policies and training in place, courts have sanctioned the individual lawyer who broke the rules rather than the firm. (See Lacey v. State Farm Gen. Ins. Co. (C.D. Cal. 2025) No. 2:24-cv-05205; Johnson v. Dunn (N.D. Ala. 2025) No. 2:21-cv-1701.) Beyond managing risk, a written policy documents that you took your duty to supervise seriously.
AI tools carry different levels of risk. What matters most for preserving confidentiality is what a tool does with the data you put into it. Two categories are worth understanding: purpose-built legal tools and general-purpose models.
A tool built for legal work, like Legion, is the safest starting point for confidential client material, because the data protections are part of the product rather than something you have to configure. Legion holds a signed Zero Data Retention agreement with its primary AI providers, which means prompts and outputs are processed in temporary memory and deleted immediately rather than stored or logged. By default, nothing you put into the platform is used to train any AI model. Data is encrypted at rest with AES-256 and in transit with TLS 1.2 or higher, and it is stored and processed exclusively on servers within the United States.
Legion runs on the same cloud infrastructure trusted by banks and hospitals, and its core service providers maintain SOC 2 Type II certifications. Because it is built specifically for litigators, the workflows, formatting, and safeguards a litigator needs are already baked in. A licensed attorney still reviews and remains responsible for every document before it is used, but it is the best starting place. (For a real-world example, see how a San Francisco firm used AI drafting to cut staffing costs and turnaround time.)
General-Purpose Models Can Be Safe Too
You do not have to use a legal-specific tool. General-purpose models like Claude, Gemini, and ChatGPT can be used safely, and there are many good options. What matters is the type of account you use.
A common and serious mistake is letting people use personal accounts for client work. Personal and free consumer accounts generally do not carry the privacy and data-handling protections that confidentiality requires. Many retain your inputs, and some reserve broad rights to use that data to train their models. Putting client information into a personal account can amount to handing it to a third party, which is exactly what your confidentiality duty forbids.
The rule for your policy is simple. Use a business or enterprise account. Confirm that the provider does not train on your data and offers a zero-retention or limited-retention setting. And require that the business account be used only for firm work, with no personal logins for client matters, ever. A business account is what turns a consumer chatbot into a tool you can defend using.
Practice Point
Before approving any tool, check the privacy policy or ask the vendor four questions: Is data encrypted in transit and at rest? Who at the company can access it? How long is it retained? And what happens to it if the company is breached or sold? For higher-stakes vendor relationships, our guide to drafting and negotiating AI data and IP license agreements covers the contract terms to insist on.
What the California Rules Actually Require
Reassuringly, AI does not rewrite your ethical obligations; it applies familiar ones to a new tool. The State Bar makes this explicit by mapping AI use onto rules you already follow. (State Bar of Cal., Standing Comm. on Prof. Responsibility & Conduct (COPRAC), Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (Nov. 16, 2023, rev. 2026).) That guidance is advisory rather than binding, but it is the clearest roadmap the Bar has offered, and it points to rules that are binding. Keep current as the guidance evolves, but do not expect the core duties to change much.
Competence
You must provide competent representation, applying the diligence and skill reasonably necessary. (Cal. Rules of Prof. Conduct, rules 1.1, 1.3.) Applied to AI, competence means understanding a tool's capabilities and limits well enough to use it responsibly. You do not need to be an engineer. You do need to know that these models can produce fluent, confident, and entirely fabricated output. The ABA's first formal guidance on AI frames this as a duty to grasp the benefits and risks of the technology you use. (ABA Standing Comm. on Ethics & Prof. Responsibility, Formal Opn. No. 512 (July 29, 2024).)
Confidentiality
California imposes a strict confidentiality duty, obligating you to protect client secrets at every peril to yourself. (Bus. & Prof. Code, § 6068, subd. (e)(1); Cal. Rules of Prof. Conduct, rule 1.6.) You may not enter confidential client information into an AI tool that presents material risks to confidentiality or security absent your client's informed consent. (Cal. Rules of Prof. Conduct, rule 1.0.1, subd. (e).) This is the rule that turns the account question from the prior section into an ethics question, not just an IT preference.
Candor and Verification
You may not present false statements of law or fact to a court, and you may not advance frivolous contentions. (Cal. Rules of Prof. Conduct, rules 3.1, 3.3.) AI's tendency to hallucinate citations makes this the most visible risk in the rules. Lawyers who filed briefs with fabricated cases have been sanctioned, from the case that produced the first widely publicized sanctions to California's own developing case law. (Mata v. Avianca, Inc. (S.D.N.Y. 2023) 678 F.Supp.3d 443; Noland v. Land of the Free, L.P. (Cal. Ct. App. Sept. 12, 2025) No. B331918 [first published California opinion addressing AI-fabricated citations].) The safeguard has not changed since before AI existed: read and verify every authority before it goes into a filing.
Supervision
If you manage other lawyers or nonlawyer staff, you must make reasonable efforts to ensure their conduct conforms to the rules. (Cal. Rules of Prof. Conduct, rules 5.1, 5.3.) This is the duty your written policy and training satisfy. It also reaches your vendors: supervising AI use includes vetting the tools your firm relies on in the first place.
Communication and Fees
You must keep clients reasonably informed and respond to their questions, which can include explaining your use of AI when it materially affects the representation. (Cal. Rules of Prof. Conduct, rules 1.2, 1.4.) And your fees must be reasonable. (Cal. Rules of Prof. Conduct, rule 1.5; Bus. & Prof. Code, §§ 6147, 6148.) If AI lets you finish a task in one hour instead of five, you bill the one hour. You do not bill for the time the tool saved you.
What Is Coming
The duties above are settled, even as the way they are packaged continues to shift. COPRAC has approved proposed amendments that would add AI-specific comments to several Rules of Professional Conduct, including the competence, communication, confidentiality, candor, and supervision rules, stating expressly that independently verifying AI output is part of your existing duties. The public comment period closed in May 2026, and as of this writing the amendments have not been formally adopted. Separately, the Judicial Council adopted a rule governing the courts' own use of AI, effective September 1, 2025, but that rule governs court staff and judicial officers, not the private attorneys who appear before them. (Cal. Rules of Court, rule 10.430.) Keep an eye on these developments, but do not wait for them. Nothing in the pending changes asks more of you than careful practice already does.
Addressing AI in Your Engagement Letters
Two AI conversations belong in your engagement letter. The first discloses your firm's use of AI. The second addresses your client's use of AI, which is the risk most firms forget. A client who pastes your draft settlement agreement into a free chatbot to "check it," or who asks an AI for legal advice between meetings, can waive privilege or act on bad information without ever telling you. Say so up front.
The clauses below are templates. Adapt them to your practice, and run them past your own review before using them. You can download both clauses here to lift into your engagement agreement or attach as an addendum.
Addendum A – Firm's Use of AI
Use of Artificial Intelligence Tools.
To assist in providing efficient and cost-effective legal services, our firm may use artificial intelligence (AI) platforms when working on your matter, including tools designed specifically for legal professionals that assist with drafting, research, and document review.
We use AI platforms that maintain appropriate confidentiality and security protections, including encryption and data isolation. We do not input confidential client information into public AI tools that lack adequate safeguards. We will review all AI-generated output for accuracy and completeness, and we will independently verify any legal authorities before relying on them in your matter.
We will bill only for the actual time we spend using AI tools and reviewing their output, not for time hypothetically saved. Some AI tools carry third-party costs, which we will disclose to you and may charge to your matter as a cost.
By signing this agreement, you acknowledge that you have read and understand the above and consent to our firm's use of AI tools as described.
Example – per-page AI tools, such as Legion:
Our firm uses Legion, an AI drafting platform built for litigators that bills on a per-page basis. Because Legion can complete in hours work that would traditionally take days or weeks, it often produces significant cost savings and better results for you. These per-page charges are billed to your matter as a cost and itemized on a monthly invoice, so they can be recouped as a case expense.
Addendum B – Client's Use of AI
Use of Artificial Intelligence.
"AI tools" means chatbots and other programs that generate, summarize, or rewrite text for you, such as ChatGPT, Claude, Gemini, and Copilot, and the AI features now built into search engines, email, and word processors.
Why this matters for your case.
Your private information can stop being private. Normally, what you share with us is protected by something called the attorney-client privilege, a rule that keeps our communications confidential and out of the other side's hands. If you type details about your case into a public AI tool, you may be handing that information to an outside company, and that can destroy the protection. Once it is gone, the other side may be entitled to demand and read what you shared. (Lawyers call information the other side can force you to turn over "discoverable.")
The AI can simply be wrong. These tools regularly produce answers that are incomplete, misleading, or entirely made up, including fake court cases that look real. Relying on them can give you a false picture of your case and cost extra time and money to untangle.
It can get you in trouble with the court. Using AI in your case can violate court rules and orders, which can lead to penalties.
What we ask of you.
Keep case information out of AI tools. Don't enter documents, our communications, or anything you've received in the case.
Don't present AI's work as your own. If you used AI to write or research something you send us, tell us.
Tell us if you've used AI at all in connection with your case, so we can address any problems early.
Rely on us, not the AI, for legal advice. We are your lawyers; the AI is not, and it should not guide decisions about your matter.
How we use AI.
We may use AI tools ourselves, for research, drafting, and review, but always with the safeguards described above, and always with a licensed attorney checking the work before we rely on it or send it to you.
If these rules aren't followed.
You could lose the privacy protections on your information, face penalties from the court, run up extra fees to fix the problem, and, in serious or repeated cases, we may have to stop representing you.
Bringing It Together
Adopting AI safely is mainly a management problem with a technology component. Firms that get this right bring AI use into the open and write down the rules. They choose tools, whether purpose-built like Legion or a general-purpose model on a business account, based on what happens to client data, and they treat AI the way the ethics rules already tell them to treat everything else: with competence, confidentiality, supervision, and candor.
Key Takeaways
A "no AI" policy does not stop AI use; it hides it. Start with an honest inventory and a short written policy. Never let confidential client information into a personal or free consumer account: require a business account with no-training and zero- or limited-retention terms, or use a purpose-built legal tool. The California ethics duties have not changed, and the most important one is unchanged from the paper era. Verify every citation and every fact before it leaves your office.
Want a tool that's safe for legal work out of the box?
Legion is purpose-built for litigators. Zero Data Retention, no training on your data, AES-256 encryption, and SOC 2 Type II infrastructure are baked into the product, so the confidentiality safeguards are there by default rather than something you have to configure. A licensed attorney stays in control of every document.
Get started with Legion →Disclaimer. This article is provided for general informational purposes only and does not constitute legal, ethical, or professional advice, and it does not create an attorney-client relationship. It is not a substitute for advice from a licensed attorney in your jurisdiction. The sample engagement-letter clauses are templates that must be reviewed and adapted to your practice, your clients, and your jurisdiction before use. The law in this area is evolving: the State Bar's generative-AI guidance is advisory, the related Rules of Professional Conduct amendments were pending as of mid-2026, and cited authorities may be superseded. Confirm the current rules, statutes, and case law before relying on anything here. Legion is a legal-technology platform, not a law firm, and does not provide legal advice. References to specific tools, pricing, or providers are illustrative and subject to change.
