The New Normal: AI in the Legal Practice
Artificial intelligence (“AI”) has become part of the modern patent litigation workflow for many attorneys.[1] It can be used to assist with various tasks such as summarizing technical documents, organizing discovery, identifying prior art, analyzing claim language, assisting in drafting expert reports, finding case law for briefs, and filtering through massive document productions for relevant information and/or document review projects. Many users have chosen their preferred AI platform, becoming familiar with its capabilities, limitations, and how to tailor their prompts to obtain the most useful results. Some of the heavy hitter AI models include “Harvey,” Anthropic’s “Claude,” Google’s “Gemini,” and of course, OpenAI’s “ChatGPT.”[2] Many law firms have also partnered with AI providers to develop enterprise platforms that incorporate enhanced security measures, including encryption, contractual confidentiality protections, and restrictions on the use of client data to train AI models, allowing attorneys to use these tools with sensitive or confidential information while reducing the risks associated with public-facing AI platforms.[3]

 

Unlike traditional legal research or document-management tools, generative AI generally operates through an interactive exchange: a user provides the AI with prompts, instructions, documents, or other information, and the AI processes those inputs to generate a responsive output.[4] In the litigation context, that exchange may require attorneys and experts to input case-specific facts, general litigation strategies, draft analyses, or other materials that traditionally would have remained within the litigation team—raising new questions about whether those inputs, and the AI-generated outputs they produce, retain the same protections from discovery—such as attorney client privilege and the work product doctrine—as they would if an attorney had drafted them.

 

Recent decisions suggest that courts are not creating AI-specific discovery rules. Instead, they are applying familiar doctrines that were created as early as the 1500’s—including attorney-client privilege, work-product protection, and expert discovery—to work product created with the assistance of AI. The exercise is a familiar one: much as courts apply constitutional principles to modern issues the founding fathers could never have envisioned, courts are now applying longstanding privilege and discovery doctrines to generative AI. Understanding how courts have approached these new questions begins with understanding how each of these longstanding protections apply.

 

Attorney-client privilege protects confidential communications between attorneys and their clients made for the purpose of seeking or providing legal advice.[5] By contrast, the work-product doctrine protects materials prepared in anticipation of litigation, including mental impressions, legal theories, and litigation strategy, and is not limited to materials personally created by attorneys.[6] Because attorney-client privilege and the work product doctrine protect different types of information under different legal standards, the same interaction with an AI tool may be protected under one doctrine but not the other. Although the case law remains limited thus far, the two cases outlined below offer an early framework for litigants navigating AI-assisted practice.

 

A Third Wheel in the Attorney-Client Relationship: Does a Client’s Independent Use of Generative AI Destroy Attorney Client Privilege?
The Southern District of New York recently considered whether a criminal defendant’s independent use of Anthropic’s Claude platform to assess his potential legal exposure, defenses, and litigation strategy gave rise to attorney-client privilege or work-product protection in United States v. Heppner.[7] While represented by counsel and without counsel’s knowledge or direction, Heppner used Claude to generate documents analyzing the government’s investigation and potential charges, including information he had obtained from his attorneys.[8]

 

The court concluded that the resulting AI communications were not privileged attorney-client communications because they were not made between a client and his attorney, “[b]ecause Claude is not an attorney.”[9] Further, these communications were not intended to be nor kept confidential since Heppner communicated with a third-party platform.[10] Nor were these communications found to be protected by the work product doctrine. The judge decided that the AI documents were neither prepared by or at the behest of counsel and therefore not reflective of counsel’s strategy.[11]

 

For in-house counsel, Heppner underscores the risks associated with employees independently using public-facing AI platforms to evaluate legal disputes or analyze advice received from counsel. Companies should consider establishing clear protocols governing the use of AI for legal matters and caution employees—particularly potential fact witnesses who might be working with outside counsel for litigation—against entering privileged communications or sensitive litigation information into unapproved AI tools without first consulting counsel.

 

Showing Your Work: Must Experts Disclose Their AI Prompts?
Current questions surrounding AI extend beyond privilege; they also run into the realm of expert discovery. In Conservation Law Foundation, Inc., a magistrate judge in the District Court of Connecticut recently addressed whether an expert witness’s prompts to a generative AI platform were discoverable as part of the expert’s methodology.[12] The case arose from an environmental lawsuit concerning Shell’s operation of a petroleum storage terminal in Connecticut.[13] The plaintiff’s expert and her research team used generative AI to sift through thousands of documents produced in discovery and identify a smaller subset of potentially relevant materials for her expert analysis.[14] Shell subsequently sought discovery of the prompts and queries used in that process, arguing that they were relevant to understanding how the expert selected the documents underlying her opinions.[15]

 

The magistrate judge agreed, concluding that AI prompts and queries used to narrow the document production were part of the expert’s methodology and therefore subject to discovery under Federal Rule of Civil Procedure 26(b).[16] The magistrate judge’s decision has since been stayed pending consideration of the plaintiff’s Rule 72(a) objection, leaving the ultimate discoverability of the prompts unresolved for the time being.[17]

 

Conservation Law Foundation highlights a potentially significant consequence of incorporating AI into expert analyses: prompts and queries used to identify or evaluate evidence may themselves become subjects of discovery. This concern is particularly relevant in patent litigation, where experts may themselves become subjects of discovery and may use AI to identify prior art, review source code, analyze technical documents, or develop infringement and invalidity opinions. Counsel should consider establishing expectations for expert AI use at the outset of litigation, including how AI-assisted analyses will be documented and independently validated, whether prompts and outputs must be preserved, and whether expert-discovery agreements should expressly address AI-generated materials.

 

Moving Forward: How to Navigate AI and Privilege Issues
For in-house counsel and their litigation teams, these decisions underscore the importance of establishing clear guidelines for the use of generative AI in litigation. Although enterprise AI platforms may offer enhanced confidentiality protections, those safeguards do not necessarily establish that communications with AI are privileged or that AI-generated materials constitute protected work product. Counsel should therefore carefully consider what information is provided to AI tools, particularly where prompts may disclose privileged communications, litigation strategy, or counsel’s mental impressions. Companies should also evaluate their AI providers’ data retention policies, confidentiality obligations, and use of customer information for model training, and distinguish between public-facing tools and enterprise platforms with contractual protections. As generative AI becomes a routine part of legal practice, the challenge for litigants will be to take advantage of its efficiencies without inadvertently expanding the scope of discoverable materials or compromising existing protections.

ABOUT BAKER BOTTS L.L.P.
Baker Botts is an international law firm whose lawyers practice throughout a network of offices around the globe. Based on our experience and knowledge of our clients' industries, we are recognized as a leading firm in the energy, technology and life sciences sectors. Since 1840, we have provided creative and effective legal solutions for our clients while demonstrating an unrelenting commitment to excellence. For more information, please visit bakerbotts.com.