Generative AI, Privilege, and Work Product: What is Protected and What is Not?

WAI CONTENT TEAM
1 day ago
5 min read

"WAI Legal Insights Blog" cover. Features a person in a circular frame, grayscale. Text on legal topics. Yellow accents, professional vibe.

By Dina Blikshteyn

Since the release of ChatGPT in November 2022, generative AI systems have become embedded in legal practice. Lawyers now use AI to draft contracts, analyze regulatory exposure, summarize records, test arguments, prepare patent applications, and develop litigation and compliance strategies. As AI becomes routine, the central question is not whether it can be used, but how traditional doctrines of attorney-client privilege and work product apply to AI-assisted communications and work products.

This article is written by Dina Blikshteyn, a partner and Co-Chair of AI and Deep Learning Practice Group and Haynes Boone, LLP. Dina focuses her practice on intellectual property portfolio management, post-grant proceedings, and AI Governance. Dina also regularly speaks and publishes on issues related to AI.

The Southern District of New York addressed part of this issue in United States v. Heppner, holding that a criminal defendant’s interactions with a publicly available AI platform were not privileged. The court did not create a categorical rule against AI-assisted privilege. Instead, it reaffirmed that traditional privilege and work product principles apply regardless of the technology involved. Protection remains a fact-specific inquiry that turns on who used the AI tool, which tool was used, and for what purpose.

Attorney-Client Privilege

Attorney-client privilege applies when communications satisfy the three elements articulated by the Second Circuit in United States v. Mejia:

a communication between attorney and client,
intended and actually kept confidential, and
for the purpose of obtaining or providing legal advice.

Each element raises distinct considerations in the AI context.

1. Communication Between Client and Attorney

Privilege protects communications between attorney and client, or with third parties engaged at the attorney’s direction to assist in providing legal advice.

As emphasized in Heppner, independent client interactions with an AI platform generally do not qualify as privileged, even if the output is later shared with counsel and incorporated into legal work. The AI platform is not legal counsel, and communications between a client and an AI platform do not satisfy the requirement that the communication be between attorney and client.

Today, clients may use AI to brainstorm, research legal concepts, analyze information, or generate draft materials. Whether those communications are protected depends on whether the AI use is integrated into the attorney-client relationship or occurs independently of it. The more detached the AI interaction is from counsel’s involvement, the weaker the privilege claim.

2. Intended and Actually Kept Confidential

Confidentiality is central to privilege. Communications must be intended to remain private and actually treated as such. In the AI context, the type of AI platform used, public or enterprise, can be decisive.

Public AI platforms. Many AI platforms operate under click-through service agreements that permit retention, reuse, or disclosure of prompts and outputs. Submitting client information to such platforms may constitute disclosure to a third party, waiving privilege.

Importantly, public AI platforms can waive privilege even if inputs are not immediately visible or publicly exposed, based on the provider’s service agreements allowing retention, reuse, or disclosure. Many public AI platforms are not actively retrained on submitted prompts and are effectively frozen in time. Even so, the absence of public exposure does not necessarily preserve confidentiality. Rather the inquiry is whether the contractual and technical framework undermines the reasonable expectation of privacy in confidential information.

Enterprise AI platforms. Paid or enterprise-grade systems often provide contractual protections that restrict data retention, prohibit model training on client data, and limit third-party disclosure. These safeguards strengthen the argument that confidentiality has been maintained.

In short, confidentiality in the AI context turns less on whether information becomes public and more on whether the legal and technical arrangements preserve a reasonable expectation of privacy in the information.

3. For the Purpose of Obtaining or Providing Legal Advice

Privilege applies only when communications are made to obtain or provide legal advice. AI platforms expressly disclaim providing legal advice and cannot substitute for counsel.

Using AI independently before consulting an attorney does not make the resulting output privileged, even if it is later shared with counsel. Privilege does not attach retroactively. The communication must occur within the context of the attorney-client relationship and for the purpose of seeking or rendering legal advice.

The closer the AI-assisted activity is tied to counsel’s provision of legal advice, the stronger the privilege argument. Where AI use occurs independently of counsel, and outside the legal advisory relationship, privilege is significantly less likely to apply.

Work Product Doctrine

Separate from privilege, the work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation or other legal proceedings. The doctrine safeguards attorney analysis, mental impressions, and strategic thinking.

AI-generated material may qualify as a work product if it is created by or at the direction of counsel and reflects legal analysis or strategy in anticipation of proceedings.

Nevertheless, in Warner v. Gilbarco, Inc., the U.S. District Court for the Eastern District of Michigan held that a pro se litigant’s interactions with ChatGPT to prepare her case were protected by the work product doctrine. The court reasoned that merely using a public AI system to prepare for litigation did not waive work product protection because the disclosure was not made to an adversary or in a manner likely to reach one.

Warner illustrates that AI involvement does not automatically destroy work product protection. At the same time, AI use does not create protection where traditional requirements are absent. As with privilege, the analysis remains fact-specific in this developing area of law.

Practical Considerations and Takeaways

Generative AI does not alter the underlying legal doctrines governing privilege and work product. At the same time, care must be used by clients and counsel when using generative AI platforms..

Key takeaways include:

Independent client use of AI is generally not privileged. Later sharing the output with counsel does not retroactively create protection.
Confidentiality depends on an AI platform. Public AI platforms may waive privilege, even if data is not publicly visible, because service agreements may allow retention, reuse, or disclosure. Enterprise AI platforms with contractual safeguards better preserve confidentiality.
Timing and purpose are critical. AI use that occurs before seeking legal advice does not become privileged simply because counsel later becomes involved.
Work product focuses on direction and anticipation of proceedings. AI-assisted materials prepared at the direction of counsel and reflecting legal strategy are more likely to be protected, while independently generated materials are less certain.
The analysis is fact-specific and evolving. Courts are applying traditional doctrine to new technology without creating categorical rules.

One theme remains constant: AI is a tool that is here to stay. Its legal implications continue to develop, and courts and litigants are grappling with how to apply traditional legal doctrines to this new technology.

_____________________________________________________________

Collaborate with us!

As always, we appreciate you taking the time to read our blog post.

If you have news relevant to our global WAI community or expertise in AI and law, we invite you to contribute to the WAI Legal Insights Blog in 2026! To explore this opportunity, please contact WAI editors Silvia A. Carretta - WAI Chief Legal Officer (via LinkedIn or silvia@womeninai.co) or Dina Blikshteyn (dina@womeninai.co).

Silvia A. Carretta and Dina Blikshteyn

- Editors