AI and Discovery: A Kansas Federal Court's Evolving Approach to Protective Orders And Other Considerations for Protective Orders

ICYMI: On March 25, 2026, U.S. Magistrate Judge for the District of Kansas Angel D. Mitchell granted defendants’ motion to amend the protective order in Jeffries v. Harcros Chemicals Inc., No. 25-2352-KHV-ADM, consolidated with Tucker v. Harcros Chemicals Inc., No. 25-2569-KHV-ADM (D. Kan.), ECF No. 152. These putative class actions concern alleged harmful emissions from a chemical facility in Kansas City, Kansas. The ruling resolved a discovery dispute and subsequent amendment to a protective order; it did not decide the underlying liability claims. The decision ultimately addresses control over discovery data, rather than the accuracy of AI-generated legal work.

The existing protective order regulated AI use involving information designated “Confidential.” Defendants sought to extend those protections to all documents and information produced in discovery. The court agreed. The existing safeguards, which the court approved extending in an amended protective order, included:

• Notice to other parties of the intended AI tool and an opportunity to object.
• Use in a secure environment.
• Restrictions on using discovery information to train or improve AI, with an exception for a tool used exclusively in the action.
• Restrictions on access to information used for training.
• Destruction or deletion requirements at the conclusion of the litigation.

Defendants were arguing for the amended protective order to expand the AI provisions “to apply not only to documents and information designated as “Confidential,” but more broadly to all “Discovery Materials”—i.e., all documents and information produced in discovery should be protected from being uploaded into public or “open loop” generative artificial intelligence tools.

Plaintiffs argued that Defendants were attempting to improperly protect all discovery through an “umbrella” order, that requiring closed AI tools would increase litigation costs, that further restriction violated the First Amendment, and that the Defendants’ asserted harms if the amended protective order were not entered were speculative. The Court rejected these arguments.

Ultimately, in applying Federal Rule of Civil Procedure 26(c)’s “good-cause standard” the court granted the amendment and approved restrictions on uploading all discovery materials—including documents not marked confidential—to public or “open” AI tools, while permitting secure, closed AI tools. In granting the motion for an amended protective order, the Court focused on four categories to satisfy the good cause standard:

• Clawback and deletion concerns: Information incorporated into AI training may be practically impossible to retrieve or delete, including inadvertently produced privileged or sensitive information.
• Security risks: Defendants operated in the chemical sector, a critical infrastructure industry, making potential exposure particularly concerning.
• Privacy issues: Bulk uploads could include employee, contractor, and other personal information, raising U.S. privacy-law and GDPR concerns.
• Discovery incentives: Secure AI use could encourage fuller productions by reducing concerns about what would happen to produced information (if left to be uploaded into an open AI tool).

Curiously, however, on July 16, 2026, the Court later vacated the Amended Protective Order, pursuant to the parties’ joint request submitted directly to the Court. A closer look at the circumstances seem to indicate that such action was taken, because the only parties remaining in the case as of June of 2026 were Plaintiffs and Harco Chemicals, Inc., which “drastically changed the discovery environment” and alleviated corresponding concerns prompting the amendment to AI protective order provisions (primarily, GDPR concerns).