Short answer: None of these states publishes a rule specifically about selling firm data to AI. Their generative-AI ethics opinions still matter, because they treat saving client information into tools that train models as a confidentiality problem. Across the states we read, firm-owned operating material is the realistic scope; client matter content stays out unless ethics counsel says otherwise.
On this page
- Read this first
- California: what the guidance says
- California: what's different in California
- California: what it implies for licensing
- Florida: what the opinion says
- Florida: what's different in Florida
- Florida: what it implies for licensing
- Texas: what the opinion says
- Texas: what's different in Texas
- Texas: what it implies for licensing
- North Carolina: what the opinion says
- North Carolina: what's different in North Carolina
- North Carolina: what it implies for licensing
- D.C.: what the opinion says
- D.C.: what's different in D.C.
- D.C.: what it implies for licensing
- Oregon: what the opinion says
- Oregon: what's different in Oregon
- Oregon: what it implies for licensing
- New Jersey: what the guidelines say
- New Jersey: what's different in New Jersey
- New Jersey: what it implies for licensing
- Pennsylvania: what the opinion says
- Pennsylvania: what's different in Pennsylvania
- Pennsylvania: what it implies for licensing
- Frequently asked questions
- Sources
Read this first
None of these opinions addresses licensing records to an AI training buyer directly. They address lawyers putting client information into generative AI tools. Their reasoning about tools that store inputs or train on them is the closest published guidance, so we summarize what each says and what it implies.
Mode lists minimum team sizes of 10+ employees for legal, 15+ for accounting, medical and finance, and 50+ for all other businesses, for companies based in the US, Canada, the UK or Australia with several years of records they own (Mode). micro1 asks for 30+ employees, mostly English data, and says it is prioritizing US companies (micro1).
This is general information, not legal advice. Talk to your own attorney or compliance advisor before you share or license any records.
California: what the guidance says
The State Bar's 2026 Practical Guidance replaces its November 2023 version. It notes that generative AI products 'often utilize the information that is input by the user, including prompts and uploaded documents or resources, to further train or refine the AI' and says, as a general matter, 'a lawyer must not input any confidential information of the client into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent' (State Bar of California).
California: what's different in California
The guidance cites Business and Professions Code § 6068(e), California's statutory duty to maintain client confidences, alongside Rule 1.6 and Rule 1.8.2. It also addresses agentic AI with access to firm email, messaging, document management and client files, which is close to what a data license would expose.
California: what it implies for licensing
A license exists to give a buyer records to train on. Under this guidance, anything confidential would need informed client consent, so a California firm's realistic scope is firm-owned material with no client information.
Florida: what the opinion says
Opinion 24-1 (January 19, 2024) applies Rule 4-1.6. It says 'it is recommended that a lawyer obtain the affected client's informed consent prior to utilizing a third-party generative AI program if the utilization would involve the disclosure of any confidential information' (Florida Bar Op. 24-1).
Florida: what's different in Florida
It describes the risk of 'self-learning' AI, which 'raises the possibility that a client's information may be stored within the program and revealed in response to future inquiries by third parties.' It also says that if a program doesn't disclose confidential information to a third party, client consent isn't required under Rule 4-1.6, which is why in-house tools are treated differently.
Florida: what it implies for licensing
Training a third party's model on client records is the self-learning risk in its most direct form. A Florida firm should assume informed consent is needed for anything confidential, and that firm-owned material is the practical scope.
Texas: what the opinion says
The Professional Ethics Committee's Opinion 705 (February 2025) warns that 'self-learning' programs pose 'a risk that the confidential information a lawyer inputs to the program may be stored within the program and revealed in responses to future inquiries by third parties. That is obviously unacceptable' (Texas Opinion 705).
Texas: what's different in Texas
Texas uses Rule 1.05, which the opinion says covers both privileged information and 'all other information relating to a client or furnished by the client and acquired by the lawyer during the course of the representation.' The opinion builds on Texas Opinion 680 on cloud computing.
Texas: what it implies for licensing
A training license is designed to put records into a model. Under Opinion 705's reasoning, confidential information shouldn't go in without client consultation and consent, which leaves firm-owned operational material as the realistic scope.
North Carolina: what the opinion says
2024 FEO 1 (adopted November 1, 2024) asks whether an AI program will retain and later use what a user provides, and concludes that 'lawyers should avoid inputting client-specific information into publicly available AI resources' (NC State Bar 2024 FEO 1).
North Carolina: what's different in North Carolina
The opinion treats in-house tools on firm-owned servers differently, but warns that customized programs 'may continue to send information inputted into the firm-specific program back to the central program for additional use or training.'
North Carolina: what it implies for licensing
An AI training buyer is the opposite of an in-house tool. The opinion's caution about client-specific information points to excluding client files from any license.
D.C.: what the opinion says
Opinion 388 asks whether the product will save information the lawyer provides, and whether the lawyer's use 'will affect the answers the GAI gives to future users of the product outside of the lawyer's law firm. Affirmative answers to either of those questions signal a need for caution' (D.C. Bar Op. 388).
D.C.: what's different in D.C.
Opinion 388 speaks to training directly: many products 'are specifically designed to collect and use information received from users—which may include client confidences and secrets—for the GAI's own training and for transmission to future users of the technology.' D.C. Rule 1.6 'requires lawyers not to reveal client confidences or secrets to third parties without the client's informed consent.'
D.C.: what it implies for licensing
Training data is saved and shapes answers for users outside the firm by design, so both of Opinion 388's warning signs are present. D.C. firms should keep client confidences and secrets out of scope.
Oregon: what the opinion says
The opinion distinguishes 'open' models, which learn from inputs, from 'closed' ones. 'If lawyers would like to use their clients' information in an open model, lawyers must obtain their clients' informed consent' (Oregon Formal Op. 2025-205).
Oregon: what's different in Oregon
Oregon speaks directly to anonymization: lawyers 'must ensure that the anonymization or redaction is complete and effective, and even so, the client's informed consent may still be required.' It also says consent requires describing risks, which 'may include disclosure and the possible loss of confidentiality or valuable intellectual property rights.'
Oregon: what it implies for licensing
A buyer's de-identification doesn't settle the question in Oregon. The opinion suggests that even well-anonymized client material may need informed consent.
New Jersey: what the guidelines say
The preliminary guidelines (January 2024) quote RPC 1.6(f): a lawyer must 'make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information related to the representation of a client.' They add that 'a lawyer is responsible to ensure the security of an AI system before entering any non-public client information' (N.J. Supreme Court guidelines).
New Jersey: what's different in New Jersey
The guidance comes from the state Supreme Court rather than a bar committee, and it quotes New Jersey's RPC 1.6, under which a lawyer shall not reveal information relating to representation 'unless the client consents after consultation'.
New Jersey: what it implies for licensing
The guidelines focus on security before entering 'non-public client information'. A license that transfers such information to a third party for training would need client consent, so firm-owned material is the practical scope.
Pennsylvania: what the opinion says
The joint opinion of the Pennsylvania and Philadelphia bar associations says large language models 'may run afoul of Rules 1.7 and 1.9 by using the information developed from one representation to inform another. Therefore, a lawyer must not input any confidential information of a client into AI that lacks adequate confidentiality and security protections' (PA Joint Formal Op. 2024-200).
Pennsylvania: what's different in Pennsylvania
It frames AI risk as a conflict-of-interest problem (Rules 1.7 and 1.9), not only a confidentiality one: a model trained on one client's matter could inform work for another.
Pennsylvania: what it implies for licensing
Training a buyer's model on client files is the cross-matter use the opinion warns about. Keep client information out.
What could your company's records be worth?
Answer 6 questions. See the published range for a company like yours and which program fits. Takes about a minute.
Frequently asked questions
Does any state have a rule specifically about selling firm data to AI?
Not that we found. The published guidance addresses lawyers' use of AI tools; this page describes how its reasoning reads for a data license.
Is client consent the answer?
Rule 1.6 lets a lawyer reveal client information with the client's informed consent. Whether that is practical depends on how many clients and matters the records touch.
What about ABA guidance?
ABA Formal Opinion 512 (July 2024) covers lawyers' use of generative AI tools under the Model Rules.
Which states are covered here?
California, Florida, Texas, North Carolina, the District of Columbia, Oregon, New Jersey and Pennsylvania, based on published opinions we read.
Sources
- ABA Formal Opinion 512 on generative AI tools (July 2024)americanbar.org
- ABA Model Rule 1.6, Confidentiality of Informationamericanbar.org
- Mode / BizBuyData referral pagebizbuydata.com
- micro1 company referral pagemicro1.ai
- State Bar of California, Practical Guidance for the Use of Generative AI in the Practice of Law (2026)calbar.ca.gov
- Florida Bar Ethics Opinion 24-1 (Jan 19, 2024)floridabar.org
- Professional Ethics Committee for the State Bar of Texas, Opinion 705 (Feb 2025)tcle-web.s3.amazonaws.com
- North Carolina State Bar, 2024 Formal Ethics Opinion 1 (adopted Nov 1, 2024)ncbar.gov
- D.C. Bar Ethics Opinion 388, Attorneys' Use of Generative AI in Client Mattersdcbar.org
- Oregon State Bar Formal Opinion 2025-205, Artificial Intelligence Toolsosbar.org
- New Jersey Supreme Court, Preliminary Guidelines on the Use of AI by New Jersey Lawyers (Jan 2024)njcourts.gov
- Pennsylvania and Philadelphia Bar Associations, Joint Formal Opinion 2024-200pa.gov
Referral-fee disclosure: we may be paid by a program if your company signs with it after using this site. General information only, not legal advice.
Compare programs
Best AI data program for a small law firm (10-29 staff)Side by side from each program's own pages. Rankings follow published rules, not referral fees.
See buyers that name law and professional-services firms, from their own pages, with sources and dates.