Industry News
ABA Formal Opinion 512: What Lawyers Must Do When Using Generative AI
The ABA's first formal ethics opinion on generative AI maps the duties of competence, confidentiality, candor, billing, and supervision onto tools like ChatGPT — with no new rules, just no excuses.
ABA Formal Opinion 512, issued July 29, 2024, is the American Bar Association's first formal ethics guidance on lawyers' use of generative AI. It creates no new rules — it maps six existing duties (competence, confidentiality, communication, candor to the tribunal, supervision, and fees) onto AI-assisted practice, and warns that unverified AI output can trigger malpractice exposure or Rule 11 sanctions.
Generative AI tools such as ChatGPT, Copilot, and legal-specific research assistants moved from novelty to daily workflow inside law firms faster than most bar regulators could respond. On July 29, 2024, the American Bar Association's Standing Committee on Ethics and Professional Responsibility closed part of that gap with Formal Opinion 512, "Generative Artificial Intelligence Tools," the ABA's first formal ethics opinion addressing how the Model Rules of Professional Conduct apply when a lawyer uses generative AI (GAI) in a representation.
What is ABA Formal Opinion 512?
Formal Opinion 512 does not regulate any specific product, and it does not create new ethics rules. Instead, it applies existing Model Rules to a new category of tools. Its syllabus states that lawyers using generative AI "must fully consider their applicable ethical obligations, including their duties to provide competent legal representation, to protect client information, to communicate with clients, to supervise their employees and agents, to advance only meritorious claims and contentions, to ensure candor toward the tribunal, and to charge reasonable fees." The opinion builds on the ABA's earlier guidance on lawyers' use of technology and outsourcing, and it arrived after several state bars — including Florida, California, New York, and Texas — had already issued their own generative-AI guidance, as the ABA's own announcement of the opinion notes.
What does the opinion require on competence and confidentiality?
Under Model Rule 1.1 (competence), the opinion does not demand that lawyers become AI engineers, but it does require a "reasonable understanding of the capabilities and limitations" of any generative AI tool used on a matter, an understanding that must be refreshed as the technology and its risks evolve. The opinion is blunt about the consequence of skipping that step: filing or advising on unverified, hallucinated AI output without independent review is treated as conduct that could constitute malpractice, since generative AI tools do not understand the meaning of the text they produce and cannot substitute for a lawyer's own professional judgment.
Confidentiality, governed by Model Rule 1.6, gets the opinion's most detailed treatment. Many generative AI products are self-learning — they retain prompts and uploaded material and may resurface that material in later outputs — which creates a real risk that information relating to a representation could reach unauthorized third parties, even inside the same firm. Formal Opinion 512 states that a client's informed consent is required before a lawyer inputs information relating to the representation into such a tool, and it specifically rejects boilerplate consent buried in a standard engagement letter as adequate; the client needs enough detail about the tool and the risk to give meaningful consent. The opinion also points lawyers toward reviewing a vendor's terms of use and, where needed, consulting IT or cybersecurity specialists before adopting a tool for client work.
How does Formal Opinion 512 treat candor to the tribunal?
The opinion's candor discussion (Model Rules 3.1, 3.3, and 8.4(c)) responds directly to a wave of court sanctions triggered by fabricated AI citations. The best-known example predates the opinion itself: in Mata v. Avianca, Inc., a Southern District of New York personal-injury case, attorney Steven A. Schwartz used ChatGPT to help draft a brief that cited nonexistent cases; when he asked the chatbot to confirm the cases were real, it falsely assured him they could be found in Westlaw and LexisNexis. Judge P. Kevin Castel sanctioned Schwartz and his firm $5,000 under Rule 11 in 2023, while noting there was "nothing inherently improper" about using a reliable AI tool — the problem was filing unverified output. Formal Opinion 512 generalizes that lesson: lawyers must independently verify any AI-generated citation, quotation, or factual assertion before submitting it to a court, because the duty of candor does not bend to whatever tool produced the document.
| Duty | Model Rule(s) | What the opinion requires |
|---|---|---|
| Competence | Rule 1.1 | Reasonable, updated understanding of a GAI tool's capabilities and limitations before relying on it |
| Confidentiality | Rule 1.6 | Informed client consent before inputting representation-related information into a self-learning GAI tool; no boilerplate consent |
| Communication | Rule 1.4 | Disclose GAI use when a client asks, when client data is input into a tool, or when it affects fee reasonableness |
| Meritorious claims & candor | Rules 3.1, 3.3, 8.4(c) | Independently verify AI-generated citations and facts before filing; never let hallucinations reach the tribunal |
| Supervision | Rules 5.1, 5.3 | Firm-wide GAI use policies; training and oversight of lawyers and nonlawyers who use the tools |
| Fees | Rule 1.5 | Bill only for actual time spent; efficiency gains from GAI cannot be billed as if they did not occur |
What does Formal Opinion 512 say about billing and supervision?
On fees, Rule 1.5's reasonableness standard still controls. The opinion states plainly that a lawyer billing hourly "must bill for their actual time," so if generative AI lets a lawyer draft a pleading in fifteen minutes that used to take three hours, the client is billed for fifteen minutes plus review time — not the old three-hour baseline. The opinion also cautions that a lawyer may not charge a client for time spent learning generative AI generally, though a lawyer may bill for learning a specific tool if a client asked for that tool's use in the matter. Flat fees and contingency arrangements are not exempt: the opinion suggests a flat fee could become unreasonable under Rule 1.5 if it no longer reflects the actual value or effort involved once AI speeds up the work, a point commentators such as the Thomson Reuters Institute have flagged as the opinion's most contested section, since it arguably conflicts with the industry's broader shift toward outcome-based pricing.
On supervision, Model Rules 5.1 and 5.3 place responsibility on managerial and supervisory lawyers to set clear firm-wide policies on permissible generative AI use and to make reasonable efforts to confirm that both lawyers and nonlawyer staff — paralegals, contract attorneys, and vendors — understand and follow those policies. The opinion draws an explicit parallel to the ABA's earlier guidance on outsourcing legal work, treating a GAI vendor much like an outside contractor whose work product a supervising lawyer remains responsible for reviewing.
Is ABA Formal Opinion 512 binding, and how has it been received?
Formal Opinion 512 is advisory. ABA Model Rules only bind lawyers once a state's highest court adopts them into that state's rules of professional conduct, and the ABA opinion itself carries no independent enforcement power. In practice, though, it functions as the closest thing to a national baseline, because most states have adopted the Model Rules' core structure, so Formal Opinion 512's rule-by-rule analysis maps cleanly onto the great majority of state ethics codes. Analysis from the National Conference of Bar Examiners and the University of North Carolina School of Law's law library both describe the opinion as foundational but preliminary, noting that many of its determinations — how much verification is enough, what informed consent must specifically say — are fact-specific and will keep getting refined as courts, state bars, and firms gain more experience with the technology. Coverage in the ABA's own Business Law Today reaches a similar conclusion: the opinion offers a useful, durable framework precisely because it ties GAI conduct to duties lawyers already owed, rather than trying to regulate a fast-moving technology rule by rule.
For firms building internal AI policy, the practical takeaway is that Formal Opinion 512 does not require avoiding generative AI — it requires treating it the way a careful lawyer would treat any powerful, imperfect research or drafting assistant: verify its output, protect client confidences going into it, disclose its use where the rules require disclosure, supervise the people who operate it, and bill for time actually worked rather than the time a machine saved.
Frequently asked
What is ABA Formal Opinion 512?
Formal Opinion 512 is the American Bar Association's first formal ethics opinion on lawyers' use of generative AI, issued July 29, 2024 by the Standing Committee on Ethics and Professional Responsibility. It does not create new rules; it applies six existing duties under the Model Rules of Professional Conduct — competence, confidentiality, communication, candor to the tribunal, supervision, and reasonable fees — to lawyers who use generative AI tools such as ChatGPT or AI-assisted legal research products in client matters. The opinion is advisory rather than binding law, but it functions as a de facto national baseline because most states have adopted the Model Rules' core structure.
Does Formal Opinion 512 require lawyers to disclose their use of generative AI to clients?
It depends on context, under Model Rule 1.4. The opinion says disclosure becomes necessary if a client specifically asks whether AI was used, if the lawyer plans to input the client's confidential information into a generative AI tool, or if AI use materially affects the reasonableness of the fee charged. It does not require blanket disclosure on every matter regardless of circumstances. Many firms address this by adding specific generative AI terms to engagement letters up front, rather than waiting for the question to arise mid-representation, which the opinion treats as good practice though not a strict mandate.
Can a lawyer get client consent to use generative AI just by adding boilerplate language to an engagement letter?
No. Formal Opinion 512 is explicit that generic, boilerplate consent language buried in a standard engagement letter does not satisfy the informed-consent requirement under Model Rule 1.6 when a lawyer plans to input information relating to the representation into a self-learning generative AI tool. The client needs enough specific detail — about the tool, what data it will process, and the risk that a self-learning system could later resurface that data — to give meaningful, informed consent. Firms are advised to review vendor terms of use and, where needed, consult IT or cybersecurity specialists before adopting a tool for client work.
How does Formal Opinion 512 address AI-generated fake case citations?
The opinion's candor discussion, tied to Model Rules 3.1, 3.3, and 8.4(c), responds to real incidents where generative AI fabricated citations that lawyers then filed with courts — most notably Mata v. Avianca, Inc., where a ChatGPT-drafted brief cited nonexistent cases and the attorney was sanctioned $5,000 under Rule 11 in 2023. Formal Opinion 512 makes clear that a lawyer's duty of candor toward the tribunal does not change because AI produced the content: every AI-generated citation, quotation, or factual claim must be independently verified before it reaches a court filing, and unverified hallucinations can support a finding of frivolous argument or lack of candor.
How should lawyers bill clients when generative AI speeds up their work?
Formal Opinion 512 holds that a lawyer billing hourly under Model Rule 1.5 must bill for actual time spent, not for the time a task would have taken without generative AI. If a tool lets a lawyer draft a pleading in fifteen minutes that previously took three hours, the client is billed for the fifteen minutes plus review time, not the old baseline. Lawyers generally may not charge clients for time spent learning generative AI in general, though they may bill for learning a specific tool if a client requested that tool's use. Flat fees are not exempt: a flat fee can become unreasonable if it no longer reflects the actual effort once AI speeds the work.
What must supervising and managerial lawyers do under Formal Opinion 512?
Under Model Rules 5.1 and 5.3, managerial lawyers must establish clear, firm-wide policies on permissible generative AI use, and supervisory lawyers must make reasonable efforts to confirm that both lawyers and nonlawyer staff — including paralegals and outside vendors — understand and follow those policies. The opinion draws an explicit parallel to the ABA's earlier guidance on outsourcing legal work, treating a generative AI vendor much like an outside contractor whose output the supervising lawyer remains responsible for reviewing. Training on the tool's capabilities, limitations, and confidentiality risks is treated as part of that supervisory duty, not an optional add-on.