Unreported / Non-Citable
Background
James H. Baqleh, who is representing himself, brought this case in the Northern District of California after his September 20, 2024 arrest. In June 2026 the parties stipulated to a protective order that mirrored the court's model order, and the court entered it. Paragraph 7.2 limits disclosure of confidential material to identified recipients unless the court orders otherwise or the designating party permits disclosure in writing. Defendant Juan Altamirano produced body-worn camera footage in discovery and marked it confidential.
Plaintiff moved to modify the protective order so he could use ChatGPT to create timestamps, prepare a chronology, transcribe the video, and identify events for discovery, motion practice, settlement, and trial preparation. He said he would not offer the resulting analysis as admissible evidence. After Defendant opposed, Plaintiff's reply narrowed the request to AI-assisted review through a business or private AI service with safeguards required by the court. Defendant argued that the footage is only about 30 minutes long and that uploading it to a public large language model risked exposing non-party criminal history, parole and probation status, drug use, police investigations, and driver's license numbers. Plaintiff acknowledged that this information is sensitive.
The Court’s Holding
Judge Haywood S. Gilliam, Jr. denied the motion because Plaintiff did not show good cause. A court may modify a protective order it has entered on good cause, which requires balancing the need for discovery against the need for confidentiality, and the party seeking relief must show that specific harm will result if the motion is denied. The court acknowledged that some courts have allowed AI-assisted review where the parties agreed to it for analyzing large quantities of data, but explained that Plaintiff was not seeking that. He instead asked permission to disclose confidential information to a third party, which the court construed as a request under paragraph 7.2.
The court found it undisputed that exposing confidential information to ChatGPT creates at least some risk about how the information will be used and safeguarded. Plaintiff proposed safeguards but cited nothing confirming that they are available or reliable. Against those risks he showed no substantial offsetting benefit. Defendant's representation that the video runs about 30 minutes went uncontested, and the court concluded that reviewing it is not an overbearing task that disadvantages a pro se party. A chronology, for example, can be made with handwritten notes.
The court said the narrowed request for a business or private AI service did not change its analysis, because the balance of some risks against essentially no benefit stayed the same. It also declined, in its discretion, to modify the order under the framework in Morgan v. V2X, Inc., a District of Colorado decision Plaintiff offered as an alternative. The order simply denies the motion; the existing protective order remains in place.
Key Takeaways
- A party who wants to send designated confidential discovery to a public AI service must show good cause, including specific harm if the request is denied.
- A small volume of material, here a 30-minute video, weighs against AI assistance because manual review is not a real burden.
- Pro se status did not by itself justify relaxing the protective order when the task could be done with ordinary tools such as handwritten notes.
- Proposed safeguards need support showing that they exist and work; unsupported assurances did not offset the disclosure risk.
- Offering a narrower option, such as a business or private AI service, does not change the result when the benefit remains negligible.
Why It Matters
This is an unreported district court order, so it is not binding precedent, and it turns on a short video, an uncontested length, and unsupported safeguards. It does not hold that AI tools can never be used with confidential discovery. It also resolves no claim arising from the arrest; it addresses only a discovery-confidentiality question.
The order does show how the court will treat such requests. It distinguishes agreed AI-assisted review of large data sets, which other courts have allowed, from a unilateral request to share designated material with an outside platform. Litigants who want to use AI should seek the designating party's written permission under the protective order, or be ready to show verified safeguards and a real need, such as a large volume of material. Whether a stronger showing on those points would change the outcome remains open.