ABA (American Bar Association) Formal Opinion 512 Review

The ABA issued its first ethics opinion on the use of GENAI (Generative Artificial Intelligence).

In this week’s newsletter, we will:

1) Summarize the main points of the opinion;

2) Highlight what could have been made clearer;

3) Take a deeper dive into disclosures of use to clients and the courts; and

4) Provide some recommendations to all practitioners.

1) Summary of key points:

Lawyers using generative artificial intelligence tools must fully consider their ethical obligations, including competence, confidentiality, communication with clients, supervision of employees, advancing meritorious claims, ensuring candor toward the tribunal, and charging reasonable fees.

Competence in using generative AI tools requires lawyers to understand the capabilities and limitations of the specific technology they are using, stay informed about developments in AI, and not solely rely on AI tools for legal advice or decision-making.

Confidentiality obligations under Model Rule 1.6 require lawyers to keep client information confidential, assess the risks of using AI tools, and obtain informed consent from clients before inputting client-related information into self-learning AI tools.

Communication responsibilities under Model Rule 1.4 may require lawyers to disclose their use of AI tools to clients, especially when the tool's output influences significant decisions in the representation, affects the basis of the lawyer's fee, or is relevant to the client's objectives.

Supervisory responsibilities under Model Rules 5.1 and 5.3 mandate that managerial lawyers establish clear policies on AI tool use, while supervisory lawyers must ensure compliance with professional obligations, provide training on ethical AI use, and verify that third-party AI providers protect client information and adhere to confidentiality standards.

Lawyers must ensure that fees and expenses related to the use of GENAI tools are reasonable and communicate the basis for these charges to clients, preferably in writing, before commencing the representation.

When billing clients based on hours expended using GENAI tools, lawyers must bill for the actual time spent and not charge for more hours than were expended on the client's behalf to ensure the fee is reasonable.

The factors outlined in Rule 1.5(a) for evaluating the reasonableness of fees also apply when charging clients for GENAI tools under flat or contingent fee agreements.

Lawyers may charge clients for disbursements incurred in providing legal services, such as court reporter fees, but should not bill clients for general office overhead expenses unless disclosed otherwise to the client.

Lawyers must maintain competence in the tools they use, including GENAI technology, and should not charge clients for time spent learning about GENAI tools unless explicitly requested by the client and agreed upon in advance.

2) Need for clarity:

Granted, Opinion 512 is an ethics opinion, not a technical paper on GENAI. Nonetheless, a more accurate description of how GENAI tools typically operate would have provided a stronger foundation for the ethical guidance that follows, allowing lawyers to make more informed decisions about the use of these tools in their practice. Some of the issues we saw are:

Over-Simplification

This opinion describes GENAI as tools that generate statistically probable outputs based on large amounts of digital text. While this is true, the description may oversimplify the underlying complexity and the distinct types of generative models and their specific applications.

The opinion touches on GENAI being "prediction tools that generate a statistically probable output when prompted," but it does not explain the concept of large language models, neural networks, or the process of training these models.

“Self-learning”

The reference to "self-learning" might lead to confusion. The authors mention "self-learning" GENAI tools but do not clearly explain how this learning occurs. There is no mention of concepts like fine-tuning or the difference between training and inference. GENAI models typically do not continue learning in the same way once deployed; they generate outputs based on pre-trained knowledge unless fine-tuned with new data. Indeed, it does not accurately reflect the current state of most commercially available GENAI tools, which use static models that do not update themselves during normal operation.

Misleading impressions

The opinion notes that GENAI tools create content that seems human-crafted. This might give the impression that GENAI tools understand context and nuances as humans do, which they do not. They generate content based on patterns in data without true comprehension or reasoning.

Scope of GENAI capabilities

This opinion does not clearly distinguish between general-purpose GENAI tools (e.g., ChatGPT) and those specifically designed for legal practice. This can lead to misunderstandings about the applicability and reliability of these tools in different contexts.

Bias and limitations

The opinion mentions that GENAI outputs may be biased or inaccurate due to data quality issues. It is also worth noting that:

- GENAI models can perpetuate and amplify biases present in training data;

- GENAI outputs may not fully understand the context or nuances of a given task or industry;

- GENAI models may not replicate human critical thinking, empathy, or ethical decision-making.

Evolving nature:

While the opinion acknowledges the rapidly changing nature of GENAI technology, it does not emphasize enough how quickly these tools evolve and the constant need for updated understanding and vigilance.

3) Disclosure to Clients and Courts

A. Discretion in Disclosure to Clients

The opinion recognizes that the need to disclose the use of GENAI tools can vary significantly depending on the specific circumstances of each case. The facts of each situation will determine whether disclosure is necessary.

Disclosure is required when GAI tools impact the representation. If the output of a GENAI tool influences a significant decision or aspect of the representation, such as predicting litigation outcomes or drafting critical documents, the client needs to be informed. If the GENAI tool's output is used to make significant decisions, such as legal strategy or client advice, disclosure is necessary. Clients have a right to know if their lawyer is relying on GENAI tools for these purposes.

Conversely, the opinion suggests that not all uses of GENAI require detailed disclosure if the tools are used in a manner that is routine, does not involve sensitive client information, and does not affect the core legal judgment or decision-making process. This practical approach allows lawyers to use technology efficiently without unnecessary administrative burdens.

Lawyers must consider the client's expectations and the terms of the engagement agreement. If a client specifically requires disclosure of all technological tools used, or if it is stipulated in the engagement agreement or outside counsel guidelines, the lawyer must comply. Also, if a client retained a lawyer for their expertise and personal judgment, using GENAI without disclosure could violate their expectations. In such cases, transparency is necessary.

When GENAI tools involve inputting information related to the representation that could be disclosed or accessed by unauthorized parties, informed consent from the client is required. This ensures the client understands the risks and benefits and agrees to the use of the technology. For situations requiring informed consent (like inputting confidential information into self-learning GENAI tools), merely adding general, boilerplate provisions to engagement letters is not sufficient. The consent must be truly informed, which requires a more detailed explanation and discussion with the client.

Finally, lawyers must disclose the use of GENAI tools if it affects the basis or reasonableness of the fees charged. Clients need to understand how technology is being used and how it impacts billing.

B. Communication with the Courts

Lawyers have ethical responsibilities to the courts under Rules 3.1 and 3.3. They must ensure that any information or content generated by GENAI tools and submitted to the court is accurate and not misleading.

Before submitting GAI-generated materials to a court, lawyers must carefully review them for accuracy, including verifying legal citations and factual assertions to avoid false statements. Moreover, if errors are identified in GAI-generated submissions, lawyers have a duty to correct them promptly to maintain candor towards the tribunal.

Lawyers always should consult and comply with local court rules regarding the use of GENAI tools, as some courts may require explicit disclosure of their use.

4) Recommendations to Practitioners

To gain a more comprehensive understanding of GENAI, we not only recommend reading and understanding Opinion 512 but also supplementing it by:

- If you are in-house: Creating an outside counsel guideline (OCG) that mandates disclosure of the use of Generative AI (GAI) tools by your legal counsel to ensure transparency and align with your expectations.

- If you are retained by a client: Drafting an engagement letter that discloses the scope of use of GENAI.

- Learning how GENAI models actually work!

- Drafting and distributing guidelines for evaluating GAI-generated content to your employees.

- Understanding the best practices for mitigating potential biases and limitations.

- Obtaining resources for maintaining technological competence in the evolving field of GAI.

- Encouraging collaboration with GENAI experts and technology specialists within the firm or through external partnerships to enhance understanding and effective use of GENAI tools. This can include setting up internal training sessions, knowledge-sharing forums, and cross-disciplinary teams.

For more information, please contact us at [email protected] .