Can a D.C. lawyer use generative AI like ChatGPT in client matters, and what ethics duties apply?

Plain-English summary

Opinion 388 (published April 2024) addresses lawyers' use of generative artificial intelligence (GAI) in client matters. The Committee's premise is that technology has not altered lawyers' fundamental ethical obligations, and it concludes the same holds for GAI: the tool can be a boon, but it can also be an untrustworthy assistant. A central point is that general-audience GAI products are not search engines querying a current database; they draw on a fixed training dataset and generate statistically probable text, which means they sometimes "hallucinate," fabricating things such as case names and citations.

The opinion grounds its analysis in the widely reported Mata v. Avianca sanctions, where a lawyer treated ChatGPT as a "super search engine," filed a brief with six fabricated citations, and doubled down when challenged. From competence, the opinion concludes that under Rule 1.1 a lawyer should have a reasonable and current understanding of how a GAI tool works and its dangers, limitations, and cost, and either have a reasonable basis to trust the outputs or review and validate them before relying on them, with more caution warranted for general-audience products lacking a legal track record. From confidentiality, the opinion concludes that under Rule 1.6 a lawyer should determine whether a product will expose inputs to the provider or other users, or let the lawyer's interactions affect later users' answers; an affirmative answer signals a need to use a more trustworthy product, negotiate better terms, or input only non-confidential data, and the opinion warns that even anonymized inputs may be re-linked to a client.

The opinion then works through the remaining duties. Under Rules 5.1 and 5.3, a lawyer must take reasonable measures to ensure supervised lawyers' and nonlawyers' GAI use conforms to the Rules. Under Rules 3.3 and 3.4, GAI outputs containing false statements, fabricated citations, or falsified evidence implicate candor to the tribunal and fairness, especially where a court has issued GAI-related orders. Under Rule 1.5, a lawyer billing purely by time spent may bill only for time actually spent even if GAI saves time, and may pass through out-of-pocket GAI costs only if the fee agreement allows. Under Rule 1.16(d), the opinion concludes a lawyer should consider whether specific GAI interactions belong in the client file.

In practice

Under this opinion, a D.C. lawyer may use generative AI but carries the same duties as with any technology. The opinion concludes that competence (Rule 1.1) requires a reasonable, current understanding of the specific tool plus either a reasonable basis to trust its outputs or independent review and validation, and it suggests the kind of diligence a reasonable business owner would do before a significant technology investment, including testing the tool against known work and verifying citations to laws, regulations, and decisions.

On confidentiality (Rule 1.6), the opinion concludes a lawyer should not input client confidences or secrets, including the lawyer's own mental impressions (work product), into a product that may expose them to the provider or other users or feed them back to future users, unless the lawyer is reasonably satisfied that will not happen; a zero-data-retention policy or a negotiated business agreement may resolve the concern, while "free" consumer products often will not. The opinion concludes that supervisory duties (Rules 5.1, 5.3), candor and fairness (Rules 3.3, 3.4), the reasonableness and disclosure of fees (Rule 1.5), and client-file obligations (Rule 1.16(d)) all continue to apply to GAI use.

Common questions

Q: Can I use ChatGPT to do legal research and cite what it gives me?

A: The opinion concludes general-audience GAI is not a reliable substitute for traditional research and can fabricate citations, so a lawyer must verify outputs; it points to Mata v. Avianca, where a lawyer was sanctioned for filing a brief with six fake citations generated by ChatGPT.

Q: Is it safe to put client facts into a generative AI tool?

A: The opinion concludes a lawyer should first determine whether the product will expose inputs to the provider or other users, or let the lawyer's interactions affect later users' answers; if so, the lawyer should use a more trustworthy product, negotiate better terms, or input only non-confidential information, because Rule 1.6 bars revealing client confidences and secrets without informed consent.

Q: Does anonymizing the client's information solve the confidentiality problem?

A: Not necessarily. The opinion urges caution, concluding that the more information a lawyer feeds a growing dataset, the greater the chance the GAI or its users can connect the dots to the client, especially where the representation is reflected in public dockets, news, or the lawyer's own website.

Q: If GAI makes me faster, can I bill the client for the value it added?

A: The opinion concludes that a lawyer who has agreed to bill solely by time may bill only for time actually spent, even if GAI reduces that time; the lawyer may pass through the GAI's out-of-pocket cost as an expense only if the fee agreement permits.

Q: Am I responsible for how my staff or associates use AI?

A: Yes. The opinion concludes that under Rules 5.1 and 5.3 a lawyer must take reasonable measures to ensure supervised lawyers' and nonlawyers' use of GAI conforms to the Rules, which can include vetting a product's privacy policy and requiring output verification.

Q: Do I have to keep my AI prompts and outputs in the client file?

A: The opinion concludes that, while a lawyer need not retain every datum, the lawyer should consider whether specific GAI interactions in a client matter should be retained as part of the client file that Rule 1.16(d) requires be surrendered on termination.

Background and rules framework

The opinion applies the D.C. Rules of Professional Conduct to GAI use without treating GAI as a special category. Rule 1.1 (competence) requires keeping abreast of relevant technology; Rule 1.6 (confidentiality) protects client confidences and secrets, including under Rule 1.6(f) against disclosure by others whose services the lawyer uses; Rules 5.1 and 5.3 impose supervisory duties over lawyers and nonlawyers; Rules 3.3 (candor to the tribunal) and 3.4 (fairness to opposing party and counsel) govern false statements, fabricated authority, and disobeying tribunal orders; Rule 1.5 (fees) governs reasonableness and expense disclosure; and Rule 1.16(d) governs the client file.

The opinion also draws on Rule 1.2 (scope of representation) and Rule 8.4 (misconduct), and it relies on the D.C. Court of Appeals confidentiality cases and on prior D.C. opinions addressing technology competence, hourly billing, and the client file.

Citations and references

Rules of Professional Conduct:

• D.C. RPC 1.1 / Model Rule 1.1 (competence, including relevant technology)
• D.C. RPC 1.6 / Model Rule 1.6 (confidentiality of information)
• D.C. RPC 5.1, 5.3 / Model Rules 5.1, 5.3 (supervision of lawyers and nonlawyers)
• D.C. RPC 3.3, 3.4 / Model Rules 3.3, 3.4 (candor to the tribunal; fairness)
• D.C. RPC 1.5 / Model Rule 1.5 (fees; reasonableness and expense disclosure)
• D.C. RPC 1.16 / Model Rule 1.16 (client file on termination)
• D.C. RPC 1.2, 8.4 / Model Rules 1.2, 8.4 (scope of representation; misconduct)

Cases:

• Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), sanctions for filing a brief with fabricated, AI-generated citations
• J.G. v. New York City Dept. of Education, 2024 WL 728626 (S.D.N.Y. Feb. 22, 2024), court rejected ChatGPT as support for a fee application

Other opinions cited:

• D.C. Bar Ethics Op. 371: technology competence and social media in practice
• D.C. Bar Ethics Op. 267 (1996): hourly billing and efficiency gains (quoting ABA Formal Op. 379 (1993))
• D.C. Bar Ethics Op. 333 (2005): the client file includes internal notes and memoranda

See also

• ABA Formal Op. 512: Generative AI Tools
• TX Ethics Op. 705: Generative AI in the Practice of Law
• Florida Bar Ethics Op. 24-1: Generative AI in Practice

Source

• Landing page: https://www.dcbar.org/for-lawyers/legal-ethics/ethics-opinions-210-present/ethics-opinion-388