Jeffries v. Harcros Chemicals Results in Bar on Open-Loop AI Tools
As generative AI automates everything from document summaries to deposition prep, the legal industry is grappling with an increasingly critical question of where to draw the line between efficiency and data security.
The case Morgan v. V2X, Inc. (D. Colo. Mar. 30, 2026) dealt with a related issue, when the plaintiff, appearing pro se, sought to use AI tools to bridge the technological gap with a well-funded corporate defendant. The defendant, V2X, Inc., raised alarms about whether its confidential information—including trade secrets and personnel files—was being fed into mainstream AI platforms like ChatGPT or Gemini.
Judge Maritza Dominguez Braswell’s ruling held that the identity of AI tools are not protected work product. If confidential data is being uploaded to an AI platform that may compromise its confidentiality, the opposing side has a legitimate right to know where that data is going. The concern over confidentiality is valid, with firms redoubling their efforts to keep client data private in the age of these open-source AI tools.
A recent decision from the U.S. District Court for the District of Kansas provides additional precedent in AI usage for attorneys. In Jeffries v. Harcros Chemicals Inc. (D. Kan. Mar. 25, 2026), Judge Angel Mitchell granted a motion to amend a protective order, barring litigants from uploading any discovery materials—even those not designated as confidential—into public or "open loop" AI tools.
For ediscovery professionals, this ruling highlights an evolving standard of care regarding data handling in the age of generative AI. Additionally, there’s the added importance of secure, closed-loop AI tools for the legal industry, where sensitive client data is kept private.
Limiting AI Usage Across All Discovery Materials
The litigation involves consolidated putative class actions brought by residents living near a Kansas City chemical facility, alleging chronic injuries from toxic emissions.
The court had already entered a protective order requiring parties to use closed or secure AI tools when handling confidential information. However, the defendants moved to expand these AI restrictions to cover all discovery materials produced in the case, whether marked confidential or not.
Essentially, the defendants sought a blanket ban on uploading any discovery documents into public, open AI tools. The plaintiffs pushed back, arguing that the restriction was a disfavored umbrella order that would drive up litigation costs and violate their First Amendment rights to disseminate non-confidential discovery materials.
While open loop AI tools operate in public environments and continually ingest all uploaded information to train and develop their underlying machine learning models, closed loop AI tools utilize secure environments and strict contractual safeguards that explicitly prohibit the provider from using the data to train or improve its public models. This distinction is especially important in a legal context, where extensive safeguards are put in place to protect sensitive client data.
Equal AI Restrictions and the Preservation of Public Rights
Judge Mitchell systematically dismantled the plaintiffs’ objections.
She clarified that this was not a disfavored umbrella order, as parties would still actively screen documents for confidentiality designations. Addressing the First Amendment argument, the court noted that the order does not stop parties from publicizing non-confidential documents via traditional methods (like websites or court filings); it simply controls the interim technology used to process discovery.
“A protective order does not offend the First Amendment when it is entered based on a showing of good cause, is limited to the context of pretrial civil discovery, and does not restrict the dissemination of the information if gained from other sources,” Judge Mitchell noted, citing Seattle Times Co. v. Rhinehart, 467 U.S. 20, 37 (1984).
Additionally, regarding costs, the court emphasized that the rule applies equally to both sides.
Litigants are completely free to leverage generative AI to accelerate document review, draft timelines, or summarize evidence—provided they use closed AI tools that meet basic security requirements.
Why Open AI Tools Pose a Paradigm Shift of Risk
In evaluating the arguments, Judge Mitchell found that the defendants demonstrated "good cause" under Federal Rule of Civil Procedure 26(c)(1), noting that the way AI tools function poses a unique threat to data integrity.
The court highlighted several key reasons why open AI tools are fundamentally incompatible with traditional discovery safeguards:
The Impossibility of the Claw Back: Open AI tools utilize submitted data to continually train and improve their models. The court recognized that once data is ingested by a public AI model, it is practically impossible to claw back privileged information or delete it at the conclusion of the lawsuit.
GDPR and Data Privacy Violations: Two of the defendants in the case are subject to the European General Data Protection Regulation (GDPR). The court noted that wholesale submission of discovery data to an open AI tool could expose massive amounts of personal data without the freely given, specific, informed and unambiguous consent required by GDPR Article 4(11).
Critical Infrastructure Risks: Because the defendants operate within the chemical sector—designated as critical infrastructure vital to national security—the court validated concerns that exposure of confidential data to open AI tools could hand cybercriminals a roadmap for data breaches.
The Big Takeaway for Ediscovery
By ensuring that non-confidential data won't be permanently absorbed into a public AI tool, producing parties can fulfill their discovery obligations without fear of accidental, permanent exposure.
As Judge Mitchell wrote in her ruling:
“If the court were to allow a party to upload all non-confidential documents and materials produced by another party to an open AI Tool, thereby making all such data amenable to public consumption, parties may err on the side of under-producing potentially responsive documents or seek to make extensive redactions of irrelevant or non-responsive information. Defendants’ proposed amendment allowing the use of closed AI Tools, as opposed to open AI Tools, will facilitate discovery by incentivizing more fulsome document productions.”
As data becomes increasingly complex, Jeffries v. Harcros Chemicals serves as a stark reminder for legal teams. Generative AI is a welcome evolution in ediscovery, but public, open-loop tools can be a risk in discovery. When handling sensitive data, closed and secure AI environments are often recommended to avoid risk.
