In a decision of first impression in Massachusetts, a judge in the Business Litigation Session of the Superior Court ruled that material generated by an artificial intelligence tool, or “AI output,” is not protected under the work product doctrine unless performed for or at the direction of counsel.
Shealy v. Seaside Investments, LLC, serves as a cautionary example of courts analyzing AI-generated materials through the lens of established attorney-client privilege and work product frameworks and deciding that information disclosed to a non-attorney, including an AI-powered tool, potentially exposes those materials to discovery.
The underlying dispute in Shealy v. Seaside Investments, LLC developed from a commercial investment relationship between sophisticated partners who executed a purchase agreement and promissory note in January 2025. After receiving correspondence from Seaside in May 2025 concerning a potential default under the promissory note, Earl Wade Shealy, Jr. sent Seaside’s correspondence and related transaction documents to his romantic partner, Debbi Fields, to have ChatGPT analyze and prepare a response letter. Fields then uploaded the documents to ChatGPT along with instructions or “prompts” and sent the AI-generated output back to Shealy. Notably, neither Shealy nor Fields consulted with an attorney during this process even though Shealy was represented by counsel in the transaction at issue. When the dispute escalated to litigation, Shealy produced redacted correspondence between himself and Fields, claiming that the correspondence and resulting AI output was protected as work product. Work product protection shields from discovery the mental impressions, conclusions, and legal theories that an attorney or representative prepares in anticipation of litigation.
In ruling on Seaside’s motion to compel unredacted communications, the court rejected Shealy’s contention that the AI-generated materials were protected by the work product doctrine on multiple grounds. First, the court determined that neither Fields nor Chat-GPT qualified as a “representative” within the meaning of the work product doctrine, which protects materials prepared by or for certain representatives, including attorneys, consultants, agents, and insurers. Massachusetts does not recognize romantic partners as “representatives” for purposes of work product protection, and the trial court declined to extend the doctrine to include them in this instance, ultimately concluding that exchanges between Shealy and Fields were subject to disclosure.
The court further determined that the ChatGPT tool itself was not Shealy’s “representative” for purposes of extending the work product protections to the AI output. In reaching this conclusion, the court distinguished recently-decided cases around the U.S. where certain courts have held that material prepared by ChatGPT and similar AI programs was protected as work product. For example, in Morgan v. V2X, Inc. and Warner v. Gilbarco, Inc., federal courts in Colorado and Michigan extended the work product doctrine to shield AI-generated output and prompts utilized by pro se litigants as such material reflected the mental impressions, opinions, and theories of the pro se litigants acting in the role of counsel. The Morgan court specifically noted the policy interest of leveling the playing field between represented and unrepresented parties.
However, in Shealy, the Court found those cases inapposite because Shealy was represented by counsel at the time he and Fields utilized ChatGPT to generate the output, and acted without counsel’s direction. Shealy decided entirely on his own initiative to use AI through a non-attorney third party, without consulting counsel about how the tool would be used.
The court further emphasized a critical consideration underlying its analysis, and stated: “[n]or does the output from ChatGPT constitute opinion work product. AI is a tool not a person. Nothing about what ChatGPT produces discloses the ‘mental impressions, conclusions, opinions, or legal theories of an attorney or other representative of a party concerning the litigation.’” (emphasis added). Absent attorney direction or input, the AI-generated output did not reflect an attorney’s or qualified representative’s strategic thinking or mental impressions. Consequently, the output cannot qualify for the heightened protection afforded to opinion work product. The Massachusetts trial court analogized the case to the February 2026 decision in United States v. Heppner, where a represented criminal defendant’s independent use of an AI tool without counsel’s direction fell outside the scope of privilege or work product protection and was subject to disclosure to the opposing party.
The decision in Shealy v. Seaside Investments provides important guidance as AI tools become increasingly prevalent across commercial operations and decision-making, including those related to litigation. Entities and individuals, including those represented by counsel, cannot simply outsource AI queries to non-attorneys and expect the results to be shielded from discovery. For AI tool use to receive protection, counsel must direct it; otherwise, the AI output and the prompts used to generate it may be fair game for discovery.
This ruling reflects a growing trend in which courts apply traditional privilege doctrines to AI-generated materials without modification, holding that the technology itself does not warrant departing from established frameworks. The takeaway is clear: once litigation becomes anticipated, involving counsel in any analytical work, AI-assisted or otherwise, is essential to shield a party’s mental impressions, theories, and strategies from discovery.
* * * *
If you have questions, please reach out to the author(s).
The Commercial Litigation Update blog is currently edited by Jennifer O’Connor and Adam Paine.