The ABA Standing Committee on Ethics and Professional Responsibility (the “Committee”) recently issued its first opinion on generative AI and not a day too soon. While states such as Florida, California, West Virginia, and Kentucky, as well as the District of Columbia have issued generative AI guidance, most states have yet to do the same. The states that have issued generative AI opinions have selected nearly the same exact ethics rules as relevant to the generative AI analysis as is found in the ABA’s Formal Opinion 512. The consistency in guidance provides at least some comfort in this ever-evolving component of the legal industry.
The Committee points out there is no single definition of artificial intelligence and that we have been using artificial intelligence for quite some time. Examples include electronic discovery in litigation, contracts analytics to conduct due diligence, basic research, and AI tools to help predict how judges may rule on legal questions. The opinion discusses a subset of AI which is generative, meaning it is a type of technology that creates new content. It can be used for the same purposes accomplished by more traditional AI. But generative AI (“GAI”) tools produce new text – they are prediction tools that create a “statistically probable output” when asked. To achieve this result, the generative AI tools analyze text from the internet or privately owned data sources. There are certain GAI tools designated as “self-learning,” indicating they learn from themselves as they collect more data.
Competence
Lawyers do not have to become experts to use GAI in client representation, but they must reasonably comprehend the capabilities and weaknesses of the specific GAI tool they plan to use. Lawyers can achieve the required level of competence through self-study or associating with another who has expertise in the relevant field. Notably, the Committee reminds lawyers that this is not a “static undertaking.” GAI and its applications expand every day. Lawyers are encouraged to attend continued education classes, read materials, and consult with others about GAI technology aimed at the legal industry. Lawyers must also identify innate risks with GAI—generating inaccurate output, lack of ability to comprehend, assess, or contextualize the results, and hallucinations. Use of GAI without independent verification could violate a lawyer’s duty of competence. The required degree of independent verification is fact specific, bearing in mind that lawyers will be deemed fully responsible for their work.
Confidentiality
All information relating to the representation of a client must be kept confidential absent client consent or other applicable exception. The opinion advises lawyers to evaluate the likelihood of disclosure and unauthorized access to information, the level of sensitivity of the information, difficulty in protecting the information, and barriers to client representation directly caused by the safeguards. This includes evaluating the risk of access and disclosure to those both outside and inside the firm, especially when it comes to self-learning GAI. Using GAI for client related work risks disclosure of confidential information because, in most of the generally available cost-free applications, the information you upload to it becomes part of the GAI itself (i.e. out of your control). These fact-driven assessments of the tools you are using must be made prior to inputting the information into a GAI tool.
Self-learning GAI may reveal information related to the representation to others outside of the firm that are using the same GAI tool. There is also a risk that those inside the firm who are subject to an ethical screen or who could inadvertently use the information for another client’s benefit are revealing client confidences. The Committee makes clear that informed client consent must be obtained before entering information relating to the representation into a self-learning GAI tool. Informed client consent in the form of generalized, boiler-plate language in an engagement letter won’t cut it.
However, lawyers need not obtain informed client consent, even with self-learning GAI, when there will be no input of client information related to the representation. The Committee opined that the baseline for lawyers includes reading and understanding the Terms of Use, privacy policies, and related contractual terms for any GAI tool they use to determine who will be able access the information the lawyer inputs or consult with a coworker or outside expert who has analyzed the terms. It may become necessary to consult with IT professionals or cyber security experts.
Communication
Lawyers may have a duty to communicate GAI use in a representation even when informed consent is not required. When asked, lawyers must always disclose their GAI usage to clients. Even if not questioned by the client, lawyers must discuss with clients when GAI output will influence a significant decision in the representation such as assessing jury selection. The Committee reminds lawyers that engagement agreements are the sensible place to disclose GAI use and to host client instructions.
Hallucinations, Meritorious Claims and Contentions and Candor Toward the Tribunal
As most of us have learned in the news, GAI “hallucinates” – a lot. That means it can misstate facts and law. It frequently makes cases up that it thinks should exist. Obviously, a lawyer could not submit papers with such misstatements and doing so will not be excused because GAI “did it.”
Misstatements to a court can be deemed misrepresentations under Model Rule 8.4(c) regardless of the lawyer’s intent. Accordingly, output from a GAI tool must be thoroughly examined to safeguard against false assertions being made to the court. GAI problems have already involved instances of citing fictitious opinions, erroneously analyzing authority, and using misleading arguments.
Supervision
Managerial lawyers must establish permissible use policies regarding GAI pertaining to law firm matters. Supervisory lawyers are required to make reasonable efforts to ensure the firm’s lawyer and non-lawyer staff use GAI in a manner consistent with the lawyer’s professional obligations under the ethics rules, including GAI training. The obligation to supervise also applies to supervising others outside of the law firm who are retained to use GAI tools for client representation. Third-party providers should be thoroughly evaluated before work is outsourced to them. This includes reference checks, awareness of vendor hiring practices, using confidentiality agreements, and the availability of a legal forum in the event of violations of the vendor agreement.
Fees
First and foremost, a lawyer using GAI to create work product must analyze how that use will be dealt with in billing clients. Alternative calculations of value may be appropriate, but lawyers must explain the basis of any fees and expenses prior to charging the client—including prior to charging the client for any use of GAI tools or services. Lawyers who bill their clients at an hourly rate cannot bill their clients for time they would have used on the task absent GAI when their use of GAI allowed the lawyer to complete the task in a much shorter period of time. Their actual time is what must be billed. Lawyers looking to charge GAI work as an expense should look to the principles described in ABA Formal Opinion 93-379. The Committee previously explained that overhead expenses should generally not be billed to the client absent an agreement to the contrary. Lawyers will therefore want to examine whether the use of a GAI is truly an “overhead” expense. This may be affected by issues such as whether the tool is client specific and the provider charges on a per-use basis. Firms that have created in-house GAI tools must still only charge the direct costs associated with the GAI tool and a reasonable allocation of expenses with providing the GAI tool absent a contrary agreement. Last, lawyers cannot charge time for their own inexperience, such as learning to use a GAI tool, unless the client explicitly requests the lawyer learn to use the GAI tool.