Do Not Let AI Searches Undermine Your Lawsuit

By Michael M. McMahon

Almost anyone facing legal issues such as an investigation, a lawsuit, a termination, or a contract

dispute will instinctively turn to the internet for information. With recent changes in technology,

that often means not just Google but instead turning to generative artificial intelligence (AI) tools

like ChatGPT, Claude, Gemini, or Copilot. These platforms can summarize laws, generate

timelines, draft narratives, and even assess “legal exposure” in seconds. They should never be

relied on as the final answer, but they provide a useful starting place for people with no

experience in the law. However, a potential pitfall of such efforts is potentially being forced to

turn over your AI searches if you do end up in a lawsuit.

For many people, this type of online research feels private. You are alone at your keyboard, with

no opposing party (or attorney) watching. Because you find it helpful, you plan to show the

results to your lawyer. In doing so, you may be thinking that the information will be protected

by attorney-client privilege. But recent court decisions send a stark warning that documents and

analyses generated with AI before engaging counsel may not be protected by privilege. That

means it may be subject to discovery from the other side and may ultimately be used against you

in court. [jdsupra.com], [insuranceb...essmag.com]

Attorneyclient privilege is a legal rule protecting confidential communications between the

attorney and client. The privilege is one of the most sacrosanct protections in the legal system

and has few exceptions. The privilege shields confidential communications between a client and

a lawyer made for the purpose of obtaining legal advice. While privilege matters, it does not

always apply to communication on legal matters. The privilege applies only when legal advice is

sought from a professional legal advisor in confidence, ensuring the client can speak freely to

receive effective representation.

The workproduct doctrine provides additional protection for materials prepared by, or at the

direction of, counsel in anticipation of litigation and shields it from discovery by opposing

parties. Protected work product includes documents, reports, interviews, and notes created

specifically for legal proceedings, by the attorney or at the direction of the attorney. These

doctrines exist to encourage honesty between clients and lawyers. But they are narrowly defined

and easily lost.

Generative AI sits squarely at the intersection of these doctrines. Using public generative AI

probably does not entitle the communication to protection under attorney-client privilege in

California. Recent legal developments indicate that sharing confidential information with AI

platforms is viewed as disclosing it to a third party, which destroys the confidentiality necessary

for privilege.

In United States v. Heppner, a federal judge in the Southern District of New York confronted a

question of first impression: whether a defendant’s use of a public AI tool to analyze legal

exposure and defense strategy was protected from disclosure. [mindingyou...gation.com] The

defendant, facing securities and wirefraud charges, used Anthropic’s AI platform “Claude” to

generate approximately 31 documents after learning he was the target of a criminal investigation.

He later shared those documents with his attorneys and asserted privilege when the government

sought access. The court rejected the privilege claim, completely.

The judge in the case held that the AIgenerated materials were protected by neither attorneyclient

privilege nor the workproduct doctrine, even though they were created in anticipation of

litigation, contained legal analysis and defense themes, and were later transmitted to counsel.

[venable.com], [bakerlaw.com] The ruling rested on several critical points that matter for

everyday consumers.

AI Is Not Your Lawyer

Privilege requires communication with a licensed attorney or an authorized agent of the attorney,

such as an interpreter. An AI platform, even one that sounds authoritative, is not a lawyer, does

not owe the user any fiduciary duties, and cannot form a privileged relationship.

[news.bloom...erglaw.com], [americanbar.org] As the court emphasized, all recognized

privileges are rooted in a “trusting human relationship” that cannot exist between a user and a

chatbot.

Confidentiality Was Defeated by the Platform Itself

The AI tool’s published terms allowed collection, retention, training, and disclosure of user

inputs to third parties, including government authorities. That alone defeated any reasonable

expectation of confidentiality, a core element of privilege. [jdsupra.com],

[insuranceb...essmag.com] In other words, by typing into the AI, the defendant disclosed his

thoughts to a third party.

Anticipation of Litigation Was Not Enough

Many people assume that anything created “in anticipation of litigation” is protected. That is not

true. The workproduct doctrine generally protects materials prepared by or at the direction of

counsel. In Heppner, the defendant acted independently. The court likened his AI use to a

layperson’s internet research that falls outside the doctrine’s protection. [jdsupra.com],

[dlapiper.com] Sending the documents to a lawyer later did not cure the defect. 1

Having a plan to talk to a lawyer while generating such information is one of the most common

and dangerous assumptions. Courts have long held that selfgenerated notes, timelines, and

analyses may be privileged only if they are created for the purpose of communicating with

counsel and remain confidential. AI complicates this analysis because confidentiality may be

compromised at the moment of creation. [hsfkramer.com] The court expressly rejected the

argument that AI outputs were no different from a client’s handwritten notes, the difference was

the interposed third party, namely the AI platform itself.

This is not a blanket argument against all use of AI. Rather, this is an effort to caution potential

clients about when and how to use it. Before consulting a lawyer avoid using public AI tools to

analyze legal liability, defenses, or exposure. Do not upload documents, facts, or timelines

related to a dispute and do not assume “private” means confidential.

Generative AI feels like a safe place to think through legal problems. But courts are increasingly

sending the message that doing so carries risks. And attorneys are becoming increasingly aware

that discovery aimed at obtaining a party’s AI-based searches may be useful. If you are facing a

serious legal issue, the safest first step is still the oldfashioned one: talk to a lawyer first before

you ask an algorithm.

Michael McMahon is a Partner in Carmel & Naccasha’s Litigation Practice with a focus on

business, real estate, employment, and public agency litigation, in addition to insurance defense

and corporate and business transactions. Mike can be reached at (805) 226-4148 or

mmcmahon@carnaclaw.com.

About Carmel & Naccasha 

Founded in 2004, Carmel & Naccasha has offices in San Luis Obispo and Paso Robles. The

firm’s lawyers focus their practice and provide exemplary client services in the areas of business

transactions, real property, land use, commercial and employment litigation, trusts and estate

planning, municipal law, and insurance coverage.  For more information about Carmel &

Naccasha, visit the website at www.carnaclaw.com 

Contact a Legal Professional 

The information provided herein does not, and is not intended to, constitute legal advice;

instead, all information, content, and materials are for general informational purposes only.

Neither this website nor this post is intended to create an attorney-client relationship. 

If you have any questions, please contact Carmel & Naccasha, and for more details, read our full

disclaimer. 

1 Similarly, many people mistakenly believe that they can render an existing document “privileged” simply by

sending it to their attorney.