On July 29, 2024, the American Bar Association issued Formal Opinion 512, its first formal ethics guidance on generative artificial intelligence (AI) in legal practice. The opinion answered a critical question: which Model Rules already apply when a lawyer uses an AI tool, and how? The answer covers at least five rules, including competence, confidentiality, fees, candor toward the tribunal, and supervision. It serves as the national baseline because 49 states have adopted the core structure of the ABA Model Rules. California is the outlier, but it has its own parallel guidance.
State bars did not wait for the ABA. The Florida Bar Board of Governors approved Ethics Opinion 24-1 on January 19, 2024, months before the national guidance emerged. The State Bar of California published its Practical Guidance for the Use of Generative Artificial Intelligence on November 16, 2023. The New York City Bar Association issued Formal Opinion 2024-5 on August 7, 2024, and the New York State Unified Court System adopted a new rule on AI use effective June 1, 2026. These ethical obligations are current, active, and enforceable.
This article outlines your professional duties as a practicing attorney. Whether you operate a mid-sized firm or run a practice using Legal AI for Solo & Small Law Firms: A Buyer's Guide, you must understand these legal AI ethics rules. For each duty, we examine the relevant rules, the bar opinions, and the practical translation for your daily workflow. The analysis assumes the ABA Model Rules as the baseline, so check your state bar's specific guidance for local variations.
Rule 1.1 -- Competence
According to ABA Formal Opinion 512, lawyers need not become generative AI experts. However, you must have a reasonable understanding of the capabilities and limitations of any tool you choose to use. Comment 8 to Model Rule 1.1 already requires attorneys to keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology. Opinion 512 applies this directly to AI. A competent lawyer must understand what the tool is trained on, when the training occurred, how it generates output, and why hallucinations occur. You must also know what errors are common for specific tasks and how to verify results before relying on them. This requirement is a cornerstone of lawyer AI ethics.
Other jurisdictions have established similar requirements. The State Bar of California, in its Practical Guidance, states that competence requires understanding large language models before using them. This includes grasping the specific risks of hallucinations and data privacy exposure. The California guidance emphasizes that using an AI tool without understanding its mechanics is itself a competence failure. The Florida Bar's Ethics Opinion 24-1 lists competence as one of four core ethical concerns. It reinforces that lawyers must practice accurately and competently when using AI, applying the same standard required for traditional, non-AI work.
Here is the practical translation. You must understand how your specific AI tool generates its output before using it on client matters. You must know its documented failure modes, such as hallucinated citations, mischaracterized holdings, and omitted authority. This knowledge allows you to design an effective verification workflow. To satisfy the AI competence duty lawyers face, you must maintain competence as the software updates. If a developer changes the underlying model or the knowledge cutoff date, the tool's risk profile changes, and you must evaluate it again. If you are reviewing new technologies, you can consult resources like the Best AI Legal Research Tools for Law Firms (2026) to understand different system capabilities.
Rule 1.6 -- Confidentiality
Model Rule 1.6 prohibits lawyers from disclosing client information without informed consent. Inputting client data into a generative AI tool can trigger this prohibition because the tool's data-handling processes become your disclosure problem. ABA Formal Opinion 512 recommends that lawyers secure a client's informed consent before inputting client confidences into any generative AI system. The opinion explicitly notes that boilerplate consent language in standard engagement letters is not adequate. Informed consent requires you to disclose the specific tool you are using, how that tool handles data, and what risks exist. You must evaluate the data-handling terms of any AI tool before use to confirm whether data is retained, used for model training, stored securely, or accessed by third parties.
State bars enforce strict confidentiality rules for technology. California's Practical Guidance asserts that any client-related data processed by AI must meet traditional confidentiality standards. Cloud-based systems require particular scrutiny regarding provider security protocols and data residency. The Florida Bar's Ethics Opinion 24-1 notes that client confidentiality must remain a top priority. The guidance states that confidentiality is not relaxed simply because a software vendor labels its product secure. Furthermore, the New York City Bar Association issued Formal Opinion 2025-6 to address tools that record and transcribe client meetings. This opinion mandates that lawyers obtain explicit client consent and ensure strict confidentiality before using any AI recording or transcription tool during client conversations.
Meeting your legal AI confidentiality obligations requires more than a simple vendor promise. You must review the terms of service and privacy policies of any AI tool before inputting client data. You must obtain specific, informed client consent if the tool retains, transmits, or processes client confidences, avoiding vague, boilerplate language. It is best to prefer tools that offer enterprise data agreements. These agreements must commit in writing that your data will not be retained or used to train the developer's models. Never rely solely on verbal sales claims. If a client has not consented to AI use, do not input their data into a generative system. You can use anonymized or redacted materials instead for testing or evaluation purposes. Some dedicated software options, such as those listed in our review of the Best AI Contract Review & Drafting Tools for Lawyers (2026), provide structured security options for firms.
Rule 1.5 -- Fees
Billing rules apply directly to the time spent working with new software. ABA Formal Opinion 512 states that lawyers may not charge clients for time spent learning a technology that will be used for client matters generally. There is a narrow exception. If a specific client requests that you use a particular AI tool for their matter, you may charge that client for the time spent learning how to use it. Beyond learning time, AI use has significant implications for hourly billing. If an AI tool reduces the time required to complete a task, you cannot bill for the hours you previously would have spent on manual work. Charging for pre-AI hours when a tool has dramatically compressed the actual time spent violates the requirement to charge a reasonable fee.
The Florida Bar's Ethics Opinion 24-1 aligns with this view on billing transparency. It requires lawyers to disclose AI use when it impacts client billing or costs. If an AI tool reduces the cost of a matter, the final fee must reflect that reduction. Transparency is essential so that clients understand exactly what services they are paying for and how those services were delivered.
In practice, you must not bill a client for time spent evaluating, testing, or learning an AI tool that you adopt into your general practice. You must reassess your billing model if AI substantially compresses your task times. Continuing to bill ten hours of attorney time for a task that AI helped you complete in two hours is ethically problematic. Additionally, if you pass through the cost of an AI tool subscription to a client as a disbursement, you must disclose it and ensure it reflects the actual cost incurred. You can manage these billing and administrative entries using modern solutions like the Best Legal Practice Management Software with AI (2026).
Rule 3.3 -- Candor toward the tribunal
Model Rule 3.3 requires lawyers to maintain candor toward the court. ABA Formal Opinion 512 warns that lawyers must be on guard to prevent AI hallucinations from forming the basis of frivolous claims or arguments. Submitting a brief that contains fabricated cases or mischaracterized holdings violates Rule 3.3, regardless of whether you knew the AI generated them. The opinion makes clear that you remain personally responsible for the accuracy of every citation submitted to a tribunal. Delegating citation generation to an AI tool does not transfer or reduce your verification obligation.
The real-world consequences of failing this duty are clear. In the case of Mata v. Avianca, Inc. in the Southern District of New York, lawyers submitted a ChatGPT-generated brief containing fabricated case citations. Judge P. Kevin Castel sanctioned the attorneys $5,000. The sanctions arose not just from submitting the fake citations, but from continuing to defend them after opposing counsel flagged the issue. The attorneys had submitted additional AI-generated copies of cases that did not exist. The court required the lawyers to personally notify every judge whose opinions had been cited falsely. In a later case, Johnson v. Dunn in the Northern District of Alabama, a court imposed disqualification instead of monetary sanctions when a large firm submitted hallucinated citations.
To prevent these outcomes, courts are establishing new procedural rules. Following the Mata case, federal courts in multiple districts issued standing orders. These orders require attorneys to certify that any AI-generated content used in filings has been verified for accuracy. The Northern District of Illinois, Southern District of New York, Southern District of Florida, and Eastern District of Pennsylvania are among those with active standing orders.
Your practical translation is absolute. You must independently verify every citation an AI tool produces before you file. You must check that the case exists, that it says what the AI claims, and that it has not been overruled. Never cite a case you have not personally confirmed. If a court has a standing AI certification order in your district, comply with it explicitly, even if you believe your tool is highly reliable. For an objective look at documented error rates, you can read how accurate is legal AI, really? What the benchmarks show in 2026.
Rules 5.1-5.3 -- Supervision
Supervisory duties under Model Rules 5.1, 5.2, and 5.3 apply directly to AI use within a firm. ABA Formal Opinion 512 explains that partners and managing attorneys must oversee how subordinate lawyers and non-lawyer assistants use these tools. A supervising attorney who allows a subordinate to use AI without adequate oversight remains responsible for the output. Rule 5.3 specifically applies to non-lawyer assistants, such as paralegals and legal assistants. If your staff uses generative AI to produce work product, you must ensure that their work meets the same ethical standards that apply to your own.
To maintain compliance with bar rules AI obligations, you must establish clear, firm-wide AI use protocols before staff members begin using AI tools on client matters. You must include AI output in your regular review workflow, treating it the same way you would review a junior associate's first draft. Do not assume that an AI tool's output is safe simply because a paralegal ran the prompt. You must review the draft yourself or establish a documented review step. If you are a partner or managing attorney, you are responsible for the AI-related ethical compliance of everyone you supervise. To help staff draft materials safely, firms often utilize the Best AI Legal Document Drafting Tools for Lawyers (2026), provided they maintain strict oversight workflows.
What you owe clients on disclosure
This duty synthesizes the communication obligations found in Model Rule 1.4 and the informed consent requirements of Rule 1.0. ABA Formal Opinion 512 addresses disclosure through the lens of confidentiality and fees. The opinion does not mandate a blanket disclosure of AI use in every matter. However, its structure effectively requires disclosure any time AI touches client data or client fees because you must obtain informed consent for data input and ensure billing transparency.
State bars are increasingly explicit about these requirements. The Florida Bar's Ethics Opinion 24-1 states that attorneys must disclose AI use if it impacts client billing or costs. The New York City Bar Association's Formal Opinion 2024-5 addresses communication obligations, advising attorneys to consider whether clients reasonably expect to know if AI is being used to generate work product. In California, the Practical Guidance updates emphasize that clients should be made aware, in plain language, when AI tools support their matters. This explanation should include the limitations of the technology and the risk of errors.
The practical translation is clear. If AI generates a significant portion of a deliverable, such as a research memo, contract review, or demand letter, you should consider disclosing this to the client in plain terms. Navigating AI disclosure clients expect requires a careful look at state-specific guidelines. You must disclose AI use whenever you seek informed consent for data input. If AI use changes what you bill, disclosure is required under Florida Opinion 24-1 and fits the ABA's billing principles. The overall trend across state bars is toward more disclosure. When in doubt, you should disclose your use of the technology.
FAQ
Does the ABA require lawyers to disclose AI use to clients?
ABA Formal Opinion 512 does not impose a blanket disclosure requirement for all AI use. However, it requires you to obtain informed consent before inputting client data into a generative AI tool, and to maintain billing transparency if AI affects the cost of a matter. These rules mean you must disclose AI use in most common practice scenarios. Additionally, Florida Bar Opinion 24-1 explicitly requires disclosure when AI impacts client billing or costs.
Does using AI violate confidentiality?
It does if you do not manage the data correctly. Inputting client information into a third-party AI tool is a potential disclosure of client confidences under Rule 1.6. ABA Formal Opinion 512 states that you must review the tool's data-handling terms, obtain specific informed consent instead of using boilerplate engagement language, and prefer tools with enterprise data agreements that prohibit data retention and model training.
Can I bill clients for the time I spend learning an AI tool?
No, you cannot bill for general technology training. ABA Formal Opinion 512 states that lawyers may not charge clients for time spent learning a technology adopted into their general practice. A narrow exception exists if a client explicitly requests that you use a specific AI tool for their matter, in which case you may charge for the time spent learning that tool.
What happens if AI-generated content contains fabricated citations?
You remain personally responsible for the error. In Mata v. Avianca, the court sanctioned attorneys $5,000 for submitting ChatGPT-generated briefs with fabricated case citations and continuing to defend them. Rule 3.3 does not shift responsibility to the software. The attorney who signs the filing is responsible for verifying every citation.
Do I have to verify every AI-generated citation?
Yes, you must verify all of them. ABA Formal Opinion 512 emphasizes that your duty of accuracy cannot be delegated to an AI tool. Before submitting any citation generated by AI, you must independently confirm that the case exists, holds what you claim, and remains good law. For detailed data on how often systems make these errors, you can refer to the benchmark data on legal AI accuracy.
Does my state bar have its own AI guidance?
Yes, many states have issued parallel guidelines. California published its Practical Guidance in November 2023, and Florida approved Ethics Opinion 24-1 in January 2024. The New York City Bar issued Formal Opinion 2024-5 in August 2024 and Formal Opinion 2025-6 in 2025, while New York's State Unified Court System adopted an AI rule effective June 1, 2026. While ABA Formal Opinion 512 serves as the national baseline, you must verify your specific state bar's requirements.
The bottom line
The ethical framework for AI in legal practice is not waiting for new rules to be drafted. Existing rules governing competence, confidentiality, fees, candor, and supervision already cover AI use. ABA Formal Opinion 512 has simply translated these long-standing rules into explicit, AI-specific guidance.
Your core obligations collapse to four practical requirements. First, you must understand the tool before you use it. Second, you must protect client data before you input it into any system. Third, you must verify every output before you file a document or bill a client. Fourth, you must supervise everyone at your firm who uses AI with the same rigor you apply to other professional work.
Achieving lawyer AI compliance is not about waiting for future regulations. The trend in state bar guidance is toward more explicit requirements, not fewer. Florida mandates billing disclosure, while California has proposed rule amendments requiring the verification of all AI outputs. New York's courts have already adopted an AI rule effective June 1, 2026. Staying current with your state bar's rules is a fundamental part of your competence obligation.