Legal AI sellers
Law firms will ask how your AI product keeps, protects and learns from client data, whether lawyers need client consent to use it, and how it supports supervision and fair billing, because ABA Opinion 512 and state guidance put those duties on lawyers.
Book a demoLast updated October 6, 2026
The short answer
Selling AI to law firms means answering for the lawyers' duties. ABA Formal Opinion 512 (July 29, 2024) and state guidance tell lawyers to check how a tool keeps, protects and learns from client data, get informed consent before client information goes into self-learning tools, and bill actual time. Clean finds law firms and legal teams with a real reason to buy what you sell.
Ethics rules bind lawyers, not vendors, so a law firm buying AI will ask how your product protects client information, whether it learns from the data lawyers put in, and how it supports supervision and fair fees. Have those answers in writing before the first call.
The duties start with competence. ABA Model Rule 1.1, Comment [8], says a lawyer should keep abreast of changes in the law and its practice, "including the benefits and risks associated with relevant technology." Per the LawSites tracker (opened 2026-10-06), 40 states, the District of Columbia and Puerto Rico have formally adopted that technology comment.
On top of competence sit ABA Formal Opinion 512 (July 29, 2024) and a run of state guidance. Everything on this page is context, not legal advice: read each opinion's own text, talk to your own counsel, and expect each firm to decide with its own ethics counsel.
Opinion 512 takes its vendor checklist from earlier opinions on outsourcing and applies it to generative AI providers: "reference checks and vendor credentials; understanding vendor's security policies and protocols; familiarity with vendor's hiring practices; using confidentiality agreements; understanding the vendor's conflicts check system to screen for adversity among firm clients; and the availability and accessibility of a legal forum for legal relief for violations of the vendor agreement."
On the tool itself, lawyers should make sure it is "configured to preserve the confidentiality and security of information, that the obligation is enforceable, and that the lawyer will be notified in the event of a breach or service of process regarding production of client information." They should determine whether it "retains information submitted by the lawyer before and after the discontinuation of services or asserts proprietary rights to the information." As a baseline, "all lawyers should read and understand the Terms of Use, privacy policy, and related contractual terms and policies of any GAI tool they use."
Read that as the buyer's agenda: an enforceable confidentiality commitment, a notice clause for breaches and legal demands for client data, a retention and deletion policy that covers the end of the contract, plain terms on who owns inputs and outputs. A clause letting you use customer content to improve your models will get the hardest questions.
Not for every use. Opinion 512 says "a client's informed consent is required prior to inputting information relating to the representation into" a self-learning generative AI tool. Such tools, the opinion says, "will learn from themselves as they cull more data," so one client's information could surface in answers for others. Where the lawyer will not input information relating to the representation, for example when using a tool for idea generation, the opinion says consent is not needed. Beyond those cases, it calls the confidentiality risk analysis "fact-driven".
The opinion also closes the easy route. It says "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient."
For a seller, this is the point that matters most. If your product learns from customer inputs, the firm may need informed consent before using it on client work. If it does not, say so in the contract, not only on a marketing page, and explain what happens to prompts, outputs and logs, including at any model provider. California's 2026 guidance makes the same point from the lawyer's side: "Reasonable efforts require more than reliance on generalized marketing assurances." The firm will want the clause, not the slide.
Opinion 512 puts supervision on the firm's leaders: "Managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI." It says training could cover the basics of the technology, the tools' capabilities and limitations, and secure data handling. A firm writing that policy needs your product to enforce it: who uses which features, which matters are off limits, a record of use.
Fees are the other half. The opinion says "lawyers who bill clients an hourly rate for time spent on a matter must bill for their actual time." That means time spent entering information and reviewing the draft, not the hours the tool saved. A grammar feature built into word processing software is overhead; a third-party per-use service run to review thousands of contracts for one client can ordinarily be billed as an expense at actual cost. Lawyers may also "not charge a client to learn about how to use a GAI tool or service that the lawyer will regularly use for clients." Texas Opinion 705 (February 2025) says the same about time saved.
Your pricing shapes how the firm books the cost, and California's 2026 guidance adds that subscriptions for general office AI functions typically count as overhead. What these fee rules do to your business case under hourly billing, including a Texas opinion on vendor fees tied to a firm's revenues, is covered in how to sell software to law firms.
State bars and courts mostly agree with Opinion 512 and add detail. The table is as of 2026-10-06; read each full text before relying on a line.
California deserves the closest read if you sell agentic features. On May 14, 2026, the State Bar's Board of Trustees approved updated generative AI guidance that replaces the 2023 version and, at the California Supreme Court's request, addresses agentic AI. It says lawyers "must not deploy agentic systems in a manner that allows the system to make substantive legal determinations, communicate legal advice, prepare and file pleadings, or otherwise act in a representative capacity without meaningful lawyer supervision and review." If your product takes actions on its own, show the firm where a lawyer approves each step.
The same Board item noted proposed AI amendments to the comments of the California Rules of Professional Conduct, released for public comment. On 2026-10-06 the State Bar's site still listed them in its 2026 public comment section, and we found no adoption: treat them as proposed.
State bar and court AI guidance a legal AI seller will hear about (as of 2026-10-06)
| Jurisdiction and date | What it says | What the buyer will ask you |
|---|---|---|
| California State Bar, updated guidance approved May 14, 2026 | Due diligence goes beyond marketing assurances; no agentic system acting without meaningful lawyer supervision and review; general office AI subscriptions are typically overhead | Terms and vendor documentation; approval steps for agent actions |
| Florida Bar Ethics Opinion 24-1, January 19, 2024 (advisory) | Research "data retention, data sharing, and self-learning" policies; informed consent recommended before disclosing confidential information to a third-party program; client-facing chatbots must say they are AI, not a lawyer | Retention, sharing and training policies; a chatbot disclaimer setting |
| Texas Opinion 705, February 2025 | Competence before use; protect confidential information; verify output; no charging for time saved | How users check output |
| New Jersey Supreme Court preliminary guidelines, January 24, 2024 | Ensure an AI system's security before entering non-public client information; no duty to tell clients every time, but inform a client who asks or needs to know | Security documentation before client data goes in |
| D.C. Bar Ethics Opinion 388, April 2024 | If a tool cannot be trusted with confidential information, pick another, "negotiate with the product vendor for improved confidentiality terms", or input only non-confidential data | Whether your confidentiality terms can be negotiated |
| Illinois Supreme Court AI policy, effective January 1, 2025 | AI use authorized within legal and ethical standards; disclosure "should not be required in a pleading"; review AI output before submitting it | A review step and checkable sources |
It depends on the court. The Illinois Supreme Court's policy, announced December 18, 2024 and effective January 1, 2025, says AI use "is authorized provided it complies with legal and ethical standards" and "Disclosure of AI use should not be required in a pleading," while requiring that users "thoroughly review AI-generated content before submitting it in any court proceeding." Elsewhere, a June 29, 2026 release from ILTA and the Thomson Reuters Institute said "Courts are issuing new standing orders on AI disclosure." The release gives no count.
Opinion 512 sends lawyers to the source, telling them to "consult with the applicable court's local rules to ensure that they comply with those rules with respect to AI use." Make it easy to see what the tool drafted and what a lawyer changed, so a firm can answer a disclosure order where one applies. Never imply your product removes the need to check: lawyers check each court's rules.
Most firms in ILTA's 2026 Technology Survey have a formal generative AI policy, and larger firms tend to require more. It is a single survey of 508 responding firms with more than 139,000 lawyers, released September 14, 2026; the figures describe those firms, not all firms.
Approved-app lists put the IT or innovation team between you and every lawyer; required training means you also sell the rollout. ILTA also found more than half of respondents issue licenses for the AI assistant built into their office software only on request. Who runs those steps at each size: how to sell software to law firms and the buying committee.
Clients push too. The Thomson Reuters Institute's Future of Professionals 2026 legal report is a global survey by a company that also sells legal software: 736 law firm responses in 46 countries (421 in the US) and 203 corporate legal department responses, gathered in March and April 2026. In it, 77% of clients said it is very important or essential to receive AI-enabled quality improvements from their firms, and just 5% said they get this from most or all of their providers. The same survey found 34% of law firm professionals say they use AI tools their firm has not authorized. A firm whose lawyers already use unapproved tools has a reason to approve one with controls.
Generative AI policy requirements by firm size: share of responding firms in each size band, ILTA Technology Survey 2026 (single survey, 508 firms)
| Policy requirement | Under 50 lawyers | 50 to 149 | 150 to 349 | 350 to 699 | 700 or more |
|---|---|---|---|---|---|
| Formal policy in place | 57% | 80% | 85% | 98% | 91% |
| Only vetted or firm-approved apps | 39% | 60% | 67% | 79% | 74% |
| Work product reviewed by a human | 31% | 59% | 69% | 83% | 79% |
| Training required before use | 11% | 30% | 47% | 63% | 72% |
| Client approval required | 8% | 20% | 24% | 38% | 34% |
| Supervisor or partner approval required | 4% | 10% | 13% | 19% | 19% |
Put the answers in one short document before the first call. The table is typical of what firms ask an AI vendor given the opinions above, not a list any rule requires, and each firm will add its own questions. General security material, such as certifications and questionnaires, belongs in your security review pack.
Keep one claim out of your deck: that your product satisfies Opinion 512 or has a bar's approval. These opinions tell lawyers what to check; they do not certify vendors. Say what your product does with data and let the firm conclude.
Typical AI questions from law firm buyers and what to have ready
| Question the firm will ask | Why they ask | Have ready |
|---|---|---|
| Do you keep our prompts, documents and outputs, and for how long? | Opinion 512; Florida 24-1 | Retention and deletion schedule, including at termination |
| Do you train or improve models on our data? | Opinion 512 consent for self-learning tools | A contract clause on whether customer data trains any model, yours or a provider's |
| Will you tell us about a breach or a legal demand for our data? | Opinion 512 | A notice clause with a time frame you can meet |
| Who owns inputs and outputs? | Opinion 512 | Plain ownership terms |
| Can we control who uses which features, by matter? | Firm AI policies (Opinion 512; ILTA 2026) | Admin settings by user and feature; usage logs |
| How do lawyers check the output? | Texas 705; Illinois policy | Citations in output and a review step |
| Can an agent act on its own? | California 2026 guidance on agentic AI | Where a lawyer approves each action |
| How should we treat the cost? | Opinion 512 fees; California 2026 | Pricing that separates seats from per-matter usage |
| Does the chatbot say it is an AI? (client-facing tools) | Florida 24-1 | A disclaimer setting |
The pressures on this page are also reasons firms and legal teams buy: new bar guidance, a client asking for AI value, a first AI policy, lawyers on tools nobody approved.
Clean finds law firms and legal teams with a real reason to buy what you sell, shows the evidence behind each one and names who decides. It also shows who in your team's network can introduce you. Every prospect is backed by real-world records, and Clean keeps the source behind every reason so each one can be checked. Anything Clean cannot confirm stays marked unknown.
You tell Clean who to look for by industry, size, region and what you sell, and leave out your current customers. Clean is not a list vendor, a contact database or intent data, and Clean does not send messages for you. More on legal tech buying signals, law firm technology and Clean for legal tech. Book a demo to see law firms and legal teams in your market with a real reason to buy what you sell, and who decides there.
How the tool handles client information, because the ethics duties sit with the lawyers. ABA Formal Opinion 512 (July 29, 2024) tells lawyers to understand the tool, check vendor security and terms, learn whether it keeps or claims rights to what they enter, get informed consent before representation information goes into self-learning tools, supervise use through firm policies, and bill only actual time.
Not for every use. Opinion 512 requires a client's informed consent before a lawyer inputs information relating to the representation into a self-learning generative AI tool, and it says boilerplate language in an engagement letter is not enough. Where no information relating to the representation goes in, such as generating ideas, consent is not needed. Other uses depend on the facts. Lawyers should read the opinion and their own state's guidance.
Typically: whether you keep prompts, documents and outputs and for how long; whether you train models on their data; whether you will notify them of a breach or a legal demand for client data; who owns inputs and outputs; what admin controls exist by user and matter; how lawyers check output; and whether any agent acts without lawyer approval.
It depends on the court. The Illinois Supreme Court's policy, effective January 1, 2025, says disclosure of AI use should not be required in a pleading, though users must review AI output before submitting it. Some judges issue standing orders on AI disclosure, and ABA Opinion 512 tells lawyers to consult each court's local rules. Lawyers check each court's rules.
Next
Clean