What do the ethics rules require of a lawyer who uses generative AI tools like ChatGPT in client work?
ABA Formal Opinion 512: Generative Artificial Intelligence Tools
Short answer: The opinion concludes that lawyers who use generative AI (GAI) tools must satisfy their existing duties of competence, confidentiality, communication, supervision, candor, and reasonable fees, which in practice means understanding a tool's capabilities and limits, independently verifying its output, obtaining the client's informed consent before inputting information relating to a representation, and billing only for time actually expended.
Disclaimer: This is an advisory ethics opinion. Advisory opinions are not binding; they interpret the American Bar Association's Model Rules of Professional Conduct and are persuasive authority. This summary is for research purposes only and is not legal advice. Verify current rules before acting on any specific guidance.
About this page: The plain-English summary and Q&A below were written by Ezel based on the official opinion. We do not reproduce the opinion text on this page; follow the linked source for the official text, which controls.
Plain-English summary
The opinion addresses five questions about generative AI tools: the level of competence a lawyer should acquire, how to satisfy confidentiality duties when a tool requires inputting information relating to a representation, when use of a tool must be disclosed to clients, how much review of a tool's output is necessary, and what counts as a reasonable fee or expense. It applies existing Model Rules rather than creating new ones.
On competence (Rule 1.1), the opinion states that "to competently use a GAI tool in a client representation, lawyers need not become GAI experts," but they must have a reasonable understanding of the tool's capabilities and limitations, or draw on others' expertise, and must keep up as the technology changes. Because some tools are "prone to 'hallucinations,' providing ostensibly plausible responses that have no basis in fact or reality," uncritical reliance on output without an appropriate degree of independent verification can violate Rule 1.1. The opinion adds that "regardless of the level of review the lawyer selects, the lawyer is fully responsible for the work on behalf of the client."
On confidentiality (Rule 1.6), the opinion reasons that self-learning tools can lead directly or indirectly to disclosure of information relating to a representation, including the risk that one client's information surfaces in another matter. Before inputting such information, lawyers must evaluate the fact-specific risks of disclosure inside and outside the firm, and must obtain the client's informed consent. The opinion is explicit that "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient." Consent is not required where no information relating to the representation is input.
On communication (Rule 1.4), the opinion makes disclosure of GAI use fact-specific: a lawyer must disclose if the client asks, if the engagement agreement requires it, where consent under Rule 1.6 is needed, where GAI affects the fee, or where the output will influence a significant decision in the matter. On candor and meritorious claims (Rules 3.1, 3.3, 8.4(c)), the opinion warns that GAI output, including citations to nonexistent decisions, must be carefully reviewed, and that "even an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c)." On supervision (Rules 5.1 and 5.3), managerial and supervisory lawyers must set clear policies, provide training, and oversee outside GAI providers.
In practice
Under this opinion, a lawyer using a GAI tool in client work must understand the tool well enough to use it competently and must independently verify its output to a degree appropriate to the tool and the task, remaining fully responsible for the result. Before inputting information relating to a representation, the lawyer must assess the disclosure risks and obtain the client's informed consent, which the opinion says requires explaining the risks and benefits specifically rather than relying on boilerplate. On fees (Rule 1.5), applying Formal Opinion 93-379, the opinion holds that an hourly biller may bill only for time actually expended and may not bill efficiency gains as more hours; that overhead-type tools are not separately billable while reasonable per-use costs of third-party tools may be passed through as expenses; and that, as the opinion puts it, "lawyers must remember that they may not charge clients for time necessitated by their own inexperience" in learning a tool, subject to a narrow exception where the client requests a specific tool.
Common questions
Q: Can a lawyer use ChatGPT or similar tools for client work?
A: Yes, under the opinion, provided the lawyer satisfies the duties of competence, confidentiality, communication, supervision, candor, and reasonable fees. The opinion applies existing Model Rules to GAI rather than barring its use.
Q: Does a lawyer have to independently check what the AI produces?
A: Yes. The opinion warns of "hallucinations" and says uncritical reliance without an appropriate degree of independent verification can violate the competence rule. The amount of verification depends on the tool and the task, and the lawyer remains fully responsible for the work.
Q: Do I need the client's consent before putting case information into an AI tool?
A: Per the opinion, yes, where a self-learning tool could lead to disclosure of information relating to the representation. Consent must be truly informed, and boilerplate engagement-letter language is not sufficient. No consent is required if no information relating to the representation is input.
Q: Can a lawyer bill for the hours an AI tool saved?
A: No. Applying Formal Opinion 93-379, the opinion says an hourly biller may charge only for time actually expended, so efficiency gains from a tool cannot be billed as if the lawyer spent the time.
Q: Must a lawyer tell the court when AI was used?
A: The opinion treats disclosure as fact-specific rather than automatic, but it stresses that GAI-generated citations and analysis must be reviewed before filing, and that even an unintentional misstatement to a tribunal can be a misrepresentation under Rule 8.4(c).
Background and rules framework
The opinion interprets a cluster of Model Rules as applied to generative AI: Rule 1.1 (competence, including technological competence under Comment [8]), Rule 1.6 (confidentiality), Rule 1.4 (communication), Rule 1.5 (reasonable fees and expenses), Rule 3.1 and Rule 3.3 (meritorious claims and candor to the tribunal), Rule 8.4(c) (misrepresentation), and Rules 5.1 and 5.3 (supervisory duties over lawyers and nonlawyer assistance, including outside vendors). It draws on prior technology opinions, including Formal Opinion 477R (2017) on securing client information, Formal Opinion 498 (2021) on virtual practice, and Formal Opinion 08-451 (2008) on outsourcing.
Citations and references
Rules of Professional Conduct:
- ABA Model Rule 1.1 (competence), 1.6 (confidentiality), 1.4 (communication)
- ABA Model Rule 1.5 (fees), 3.1 (meritorious claims), 3.3 (candor), 8.4(c)
- ABA Model Rule 5.1 and 5.3 (managerial and supervisory responsibilities)
Cases:
- Baker v. Humphrey, 101 U.S. 494 (1879), lawyer's duty of fidelity to the client
- Smith v. Lewis, 530 P.2d 589 (Cal. 1975), competence and legal research
Other opinions cited:
- ABA Formal Op. 93-379 (1993): billing for fees, disbursements, and expenses
- ABA Formal Op. 477R (2017): securing communication of protected client information
- ABA Formal Op. 498 (2021): virtual practice; ABA Formal Op. 08-451 (2008): outsourcing
- Florida Bar Op. 24-1 (2024) and Texas Op. 705 on generative AI in practice
See also
- FL Bar Ethics Op. 24-1: Generative AI in Practice
- TX Ethics Op. 705: Generative AI in the Practice of Law
- NYC Bar Op. 2024-5: Generative AI in NY Law Practice
Source
- Landing page: ABA Formal Ethics Opinions index
- Original PDF: aba-formal-opinion-512.pdf