AI rules for law firms and legal teams in 2026

Which AI rules apply to law firms and legal teams in 2026?

A lawyer who uses AI is bound by the rules of professional conduct of each state where the lawyer is admitted or practices, by the local rules and standing orders of each court where the lawyer files, and by ordinary privilege law. The ABA and state bar opinions below interpret those rules, and Florida’s opinion says expressly that advisory ethics opinions are not binding. What binds is the rule itself and any court order and, in California, a statute: SB 574, signed September 30, 2026, adds Business and Professions Code section 6068.1 and amends Code of Civil Procedure section 128.7, effective January 1, 2027.

A firm is also an employer and, for some tools, a provider. Its own hiring tools fall under the employment rules, and a firm that offers clients an AI product can take on vendor duties and unauthorized practice risk.

What does ABA Formal Opinion 512 require?

Opinion 512 does not add a rule. It works through the existing Model Rules one at a time and says how each applies to generative AI. The point most likely to change how a firm works is consent: the opinion says client informed consent is required before a lawyer inputs information relating to the representation into a self-learning tool whose output could disclose it, and that boilerplate consent in an engagement letter is not sufficient.

• Competence (Rule 1.1): lawyers must have a reasonable understanding of the capabilities and limitations of the tool, and the right amount of independent review depends on the tool and the task. Relying on output without an appropriate degree of review could violate the duty of competence.
• Confidentiality (Rule 1.6): beyond consent, lawyers should read the tool’s terms of use, privacy policy, and related contract terms to learn who can access what they input.
• Communication (Rule 1.4): whether to tell the client depends on the facts. Situations that call for disclosure or consultation include when the client asks, when the engagement terms or outside counsel guidelines require it, when consent is needed for inputting client information, when AI use bears on the fee, and when the tool’s output will influence a significant decision in the representation.
• Candor and meritorious claims (Rules 3.1, 3.3, 8.4(c)): before filing, lawyers must review AI output, including citations, and correct errors.
• Supervision (Rules 5.1, 5.3): managerial lawyers must establish clear policies on the firm’s permissible use and train lawyers and staff; vendor diligence covers credentials, security, confidentiality agreements, and conflicts screening.
• Fees (Rule 1.5): hourly billers bill actual time, so a lawyer may charge for the time spent prompting and reviewing a draft, not for the time saved. A flat fee unchanged by large efficiency gains may be unreasonable, and a lawyer may not bill a client to learn a tool the lawyer will use regularly.

How do state bars differ from the ABA?

They mostly agree on substance and differ on force and on consent. Of the states on this page, Florida is the one whose supreme court has already written generative AI into its rules: it added generative AI to the comments to Rules 4-1.1, 4-1.6, 4-5.1, and 4-5.3, effective October 28, 2024.

• Florida Opinion 24-1 (January 19, 2024) recommends, rather than requires, informed consent before using a third-party tool that would disclose confidential information. It says generative AI chatbots that communicate with clients or third parties must include a disclaimer that the chatbot is an AI program and not a lawyer or employee of the firm, and it bars billing that duplicates charges or inflates time.
• California approved updated Practical Guidance on May 14, 2026, which addresses agentic AI. Separately, at the California Supreme Court’s August 22, 2025 request, the Bar proposed comments to six rules, including a Rule 1.1 duty to “independently review, verify, and exercise professional judgment” over AI output and a Rule 3.3 duty to verify that cited authorities exist. According to a September 2026 report, they still needed Board of Trustees and Supreme Court approval.
• California SB 574, effective January 1, 2027, puts duties in statute rather than in bar guidance. Under Business and Professions Code section 6068.1, a California attorney may not delegate the practice of law to generative AI; may not enter confidential, personal identifying, or other nonpublic information into a generative AI system unless access is restricted to the attorney and persons the attorney authorizes under obligations to protect its confidentiality; must take reasonable steps to verify the accuracy of generative AI outputs, including all case and statutory citations, and to correct erroneous or hallucinated output in any material the attorney uses; must disclose generative AI use to the court for all documents submitted to the court; and must consider whether to disclose generative AI use in content provided to the public. The amended Code of Civil Procedure section 128.7 bars a citation in a court filing that the responsible attorney has not personally verified, whatever its source.
• Texas Opinion 705 (February 2025) says lawyers should verify any AI response and may not charge hourly fees for time “saved”; on confidentiality, it says a lawyer may need to secure client consent.
• New York City Bar Formal Opinion 2024-5 (August 7, 2024) treats consent before sharing client information as advisable and says routine uses, such as grammar checking, need not be disclosed.

What do courts require when a filing or exhibit uses AI?

Court rules split between disclosure and accountability. Either way, the signing lawyer answers for every citation. The Northern District of Texas requires a brief prepared using generative AI to say so on page one under the heading “Use of Generative Artificial Intelligence,” and treats a brief without that disclosure as a certification that no part was AI-prepared. The Illinois Supreme Court takes the other route: its policy, effective January 1, 2025, says disclosure of AI use should not be required in a pleading, while requiring all users to review AI output before filing. From January 1, 2027, California SB 574 requires attorneys to disclose generative AI use to the court for all documents submitted to the court; section 6068.1 does not prescribe the form of that disclosure. Individual judges add their own orders, so check each judge’s page before filing.

Three sanctions decisions show the range:

• Mata v. Avianca (S.D.N.Y., June 22, 2023): a $5,000 penalty, jointly and severally on two lawyers and their firm, for submitting non-existent opinions created by ChatGPT and then continuing to stand by them after court orders called their existence into question, plus letters to the client and to each judge falsely named as an author.
• Noland v. Land of the Free (Cal. Ct. App., September 12, 2025): $10,000 against counsel whose opening brief contained 23 case quotations, 21 of them fabricated, with the clerk directed to send the opinion to the State Bar. The court published the opinion as a warning that no filing should contain a citation the responsible attorney has not personally read and verified.
• Whiting v. City of Athens (6th Cir., March 13, 2026): $15,000 from each of two lawyers, plus the appellees’ attorney fees and double costs, and referral for possible discipline. The lawyers did not answer the court’s question whether they used generative AI. The court sanctioned them under Federal Rule of Appellate Procedure 38 and its inherent authority for fake citations and misrepresentations of the record, and it treated their refusal to comply with its show cause order and their prior discipline as aggravating factors.

Evidence is a separate track. For AI-generated exhibits offered today, Rule 702 and authentication under Rule 901 still govern. Proposed Federal Rule of Evidence 707 is a proposal, not a rule. It would apply the Rule 702 reliability standards to machine-generated evidence offered without an expert witness. It was published for comment in August 2025, and comment closed February 16, 2026. On May 7, 2026, the Advisory Committee on Evidence Rules did not recommend action, revised the draft, and decided not to republish it yet, choosing instead to have technology experts vet it at its fall meeting. The committee’s May report says any revised version would require republication for comment before it could go forward, and its October 15, 2026 agenda book lists artificial intelligence and proposed Rule 707.

Are AI prompts and outputs privileged?

Not automatically, and the answer turns on who used the tool, at whose direction, and on what terms. In United States v. Heppner, Judge Rakoff ruled from the bench on February 10, 2026, and explained in a February 17 memorandum, that a criminal defendant’s written exchanges with a generative AI platform were protected by neither the attorney-client privilege nor the work product doctrine. According to a Paul, Weiss summary, the court reasoned that the tool is not a lawyer, that the platform’s privacy policy reserved the right to disclose user data, and that the defendant prepared the material on his own, not at counsel’s direction.

On the same day, in Warner v. Gilbarco (E.D. Mich.), a magistrate judge held that a self-represented plaintiff’s AI-assisted litigation materials were protected work product and that using ChatGPT did not waive that protection, because work-product waiver requires disclosure to an adversary or in a way likely to reach one. Neither ruling binds other courts. Together they point to three controls: use enterprise tools whose terms keep inputs confidential, have counsel direct the use, and keep AI work inside the matter file.

When does an AI legal tool become the unauthorized practice of law?

The answer is set state by state, by statute and court rule, and how those rules apply to AI tools is being tested in litigation. The case to watch is Nippon Life Insurance Co. of America v. OpenAI Foundation (N.D. Ill., No. 1:26-cv-02448, filed March 4, 2026), in which an insurer alleges that ChatGPT engaged in the unlicensed practice of law in Illinois by helping a self-represented litigant draft filings. Those are allegations, not findings. OpenAI moved to dismiss, and the court’s August 28, 2026 minute entry stated that the motion remained under advisement. Docket entries through October 8, 2026 did not rule on it.

Marketing claims carry a separate federal risk. On February 11, 2025, the FTC announced a final order requiring DoNotPay, an AI legal services company, to pay $193,000 and barring it from claiming its service performs like a real lawyer unless it has sufficient evidence. For a firm or legal tech vendor, the defensible pattern is a licensed lawyer responsible for any advice the tool gives clients, and no performance claims the vendor cannot prove.

Does the EU AI Act or Colorado’s law reach law firms?

The EU AI Act reaches a firm as a deployer of any AI system it uses in the EU or whose output is used there, but its high-risk list does not name law firm practice. Annex III point 8(a) covers AI intended to be used by a judicial authority, or on its behalf, to research and interpret facts and law and apply the law to facts, or used in a similar way in alternative dispute resolution. Recital 61 limits the ADR part to proceedings whose outcomes produce legal effects for the parties and excludes purely ancillary administrative tools. A vendor selling such tools to courts or arbitral bodies is the provider, with high-risk duties from December 2, 2027.

For firms, the duty that applies now is to take measures to support AI literacy under Article 4, as amended in 2026. Article 50(1) binds providers, not deployers. A firm is covered only if it is the provider of a client-facing chatbot, for example because it developed the system, or had it developed, and put it into service under its own name; it must then design the chatbot so people are informed they are interacting with AI unless that is obvious. A firm that only uses a vendor’s chatbot is a deployer, and that duty falls on the chatbot’s provider. For EU hiring, deployer duties for high-risk systems listed in Annex III point 4 apply from December 2, 2027.

Colorado’s replacement law does not list legal services. SB 24-205 named “a legal service” as a consequential decision area; SB 26-189 repealed and reenacted the law with covered domains limited to education, employment, residential real estate, financial or lending services, insurance, health care services, and essential government services and public benefits. A firm doing business in Colorado can still be a deployer when it uses covered automated decision-making technology in employment decisions, with duties from January 1, 2027. A federal court order in X.AI LLC v. Weiser bars Colorado’s Attorney General from enforcing SB 26-189, including by investigation, for violations occurring on or before 14 days after the court rules on xAI’s preliminary injunction motion; xAI had not filed that motion as of October 21, 2026.

What should a contract with a legal AI vendor cover?

It should give the firm the facts its own duties require, because ABA Opinion 512 makes the lawyer, not the vendor, answerable for confidentiality and supervision. The Heppner reasoning turned in part on a consumer privacy policy, so the contract terms are part of the privilege analysis.

• No training on, retention of, or human review of client inputs and outputs without the firm’s written consent
• Confidentiality terms at least as strict as the firm’s duty under Rule 1.6, with the vendor’s security policies and breach notice
• Access controls that respect ethical walls inside the firm and isolate one client’s data from another’s
• Citation and source tracing for research and drafting features, so each authority can be checked
• A statement of what data the model was trained on and the vendor’s rights to it; in Thomson Reuters v. Ross Intelligence (3d Cir., September 29, 2026), copying Westlaw headnotes to train a legal research tool was held not to be fair use
• Logs the firm can export for litigation holds, client audits, and disciplinary inquiries
• Terms that meet clients’ outside counsel guidelines on AI, which can require disclosure under Rule 1.4

What should a firm or legal department do first?

• Inventory every AI tool lawyers and staff use, including features inside research, e-discovery, and document management platforms
• Sort tools by whether client information goes in and whether the vendor can retain or learn from it
• Write a use policy under Rules 5.1 and 5.3 and train every lawyer and assistant on it
• Replace boilerplate engagement-letter AI clauses with matter-specific informed consent where Opinion 512 requires it
• Require a human to pull and read every cited authority before any filing
• Track AI disclosure rules and standing orders for each court and judge before filing
• Before January 1, 2027, set California procedures under SB 574: no confidential client information in AI systems without restricted access, verification of every citation, and disclosure of generative AI use in documents submitted to California courts
• Bill hourly matters for actual time only, and revisit flat fees where AI changes the work
• Route client-facing chatbots and AI products through an unauthorized practice and advertising review, with an AI disclosure
• Rewrite vendor contracts for confidentiality, no training on client data, logs, and change notice

Where to read more

The AI compliance guide sets out the overall program. The EU AI Act guide covers provider and deployer duties and the Digital Omnibus dates, the US state AI laws guide covers Colorado SB 26-189, and the employment and HR guide covers a firm’s duties as an employer. The governance frameworks guide covers NIST and ISO programs that can organize the work.

This page is general information as of October 2026, not legal advice. Confirm obligations with ethics counsel, the bar of each state where you practice, and the rules of each court where you file.