Estate of Ezekiel Rivera, et al. v. Kern County Sheriff’s Department, et al. — Court enters stipulated protective order barring submission of confidential discovery material to public generative AI tools

Background

This civil action, filed February 7, 2024, with a trial date of March 7, 2028, is brought by the Estate of Ezekiel Rivera, Soluna Lora, and a minor plaintiff through his guardian against Kern County, the Kern County Sheriff's Department, two deputies, and other defendants. Pursuant to Eastern District of California Local Rule 141.1(c), the parties stipulated to a protective order, which the court entered on November 4, 2026 as Document 59.

The parties explained that discovery is likely to involve confidential or private information warranting special protection, including Kern County Sheriff's Office investigation reports, personnel records, juvenile case files, and private medical and mental-health records concerning minors. The order creates two protection tiers, 'CONFIDENTIAL' and 'HIGHLY CONFIDENTIAL – ATTORNEYS' EYES ONLY,' and limits use of protected material to prosecuting, defending, or settling this litigation.

Section 7.4 of the order, titled 'Use of Generative AI Tools,' addresses how litigants may use artificial intelligence with the protected material. It defines an 'Open Generative AI tool' as an AI system the public may access or whose underlying code and materials are publicly accessible, giving public tools such as ChatGPT as examples.

The Court’s Holding

The court ordered the stipulated protective order as modified, and Section 7.4 categorically provides that CONFIDENTIAL material or items 'shall not be submitted to any Open Generative AI tool.' The order states that providing CONFIDENTIAL information or items to an open generative AI tool is considered an unauthorized disclosure, triggering the order's remedies, which may include contempt proceedings and monetary sanctions.

The parties may submit CONFIDENTIAL material only to a 'Permitted Generative AI Tool,' defined as an enterprise AI tool that (i) restricts access to authorized users, (ii) does not use customer data to train, refine, fine-tune, or improve publicly available models, (iii) provides reasonable retention and deletion controls, and (iv) maintains commercially reasonable security and confidentiality protections. The order expressly notes that enterprise legal and productivity platforms satisfying those requirements, including Thomson Reuters CoCounsel and Microsoft 365 Copilot, are permitted.

The order otherwise confines protected material to use in this litigation, limits disclosure to enumerated categories such as counsel of record, experts, consultants, and the court, and requires that within 60 days after final disposition each receiving party return or destroy all protected material and certify compliance, subject to archival copies retained by counsel.

Key Takeaways

• Publicly accessible generative AI tools are categorically barred from receiving CONFIDENTIAL discovery material under this order; doing so is treated as an unauthorized disclosure.
• Enterprise AI tools are permitted only if they restrict access to authorized users, do not train public models on customer data, offer retention and deletion controls, and maintain commercially reasonable security and confidentiality protections.
• Named enterprise platforms such as Thomson Reuters CoCounsel and Microsoft 365 Copilot are expressly permitted when they satisfy the four criteria.
• Violations of the order, including unauthorized AI-related disclosure, may be punished by contempt proceedings and monetary sanctions.
• The order preserves each party's right to challenge confidentiality designations, seek further relief, and assert objections on grounds not addressed by the order.

Why It Matters

The order illustrates the now-common practice of embedding generative-AI restrictions directly into stipulated protective orders, converting confidentiality obligations into concrete rules about which AI systems may touch discovery material. By distinguishing open public tools from vetted enterprise platforms, the order lets litigants use AI assistance in litigation while reducing the risk that sensitive personal, juvenile, medical, and law-enforcement records are exposed to public models.

Because the AI restrictions are part of a court-entered order rather than a private agreement, violations carry the court's contempt and sanctions power, and the obligations survive final disposition until a designating party withdraws them in writing or the court directs otherwise.