Unicorn Rejects A.I. Protective/Confidentiality Order – Order Entered in Criminal Case

Unicorn Rejects A.I. Protective/Confidentiality Order – Order Entered in Criminal Case by Michael D. Berman, E-Discovery LLC.
Image: Holley Robinson, EDRM.

[EDRM Editor’s Note: The opinions and positions are those of the author.]

The Northern District of California has a “Model Stipulated Protective Order (for standard litigation)” and a “Stipulated Protective Order for Litigation Involving Patents, Highly Sensitive Confidential Information and Trade Secrets (Feb. 2022).”

I suggest that the latter was applied too stringently in Litton v. Roblox Corporation, 2026 WL 1483493 (May 27, 2026). The Litton court refused a request to modify the model order so as to prevent the use of confidential and highly confidential information in an AI tool. Id. at *4.

What happened?

Litton involves “28 related cases alleging at a general level that computerized video games designed and offered by Roblox, Microsoft, Mojang AB, and Epic Games caused harms to each of the named Plaintiffs under various legal theories, including but not limited to products liability and negligence.” Id. at *1. Three cases were designated as representative pilot cases. “Targeted discovery” related to preliminary issues was permitted.

Plaintiffs proposed four modifications to the court’s “Stipulated Protective Order for Litigation Involving Patents, Highly Sensitive Confidential Information and Trade Secrets” The court wrote that it deems the form order “presumptively reasonable and that presumption was not overcome sufficiently to warrant departing from the language in these sections of the Model Protective Order at this time.” Id. at *3. 

The presiding Magistrate Judge also issued a Standing Order for Civil Cases. In pertinent part, §VII.C states:

AI and Confidentiality: Third parties and non-parties to an action may own and operate publicly available AI tools such as large language models, machine learning tools, artificial neural networks, deep learning neural networks, and all other forms of generative AI for creating text, documents, or other materials. The owner or operator of any such AI tool may have access to, ownership of, or otherwise retain information input or submitted to such AI tool, including queries or prompts. Accordingly, in the course of preparing filings with the Court or other documents for submission in an action, counsel and Parties choosing to use an AI or other automated tools shall fully comply with any applicable protective order and all applicable ethical/legal obligations (including issues relating to privilege) in their use, disclosure to, submission to, or other interaction with any such AI tools. Such counsel and parties using any AI tool shall maintain records sufficient to establish and corroborate their compliance with this Standing Order, if asked by the Court, such as by keeping records of all prompts or inquiries submitted to any such third-party AI tools. [emphasis added].

In part of §VII.C, titled “AI and Briefs/Pleadings Filed with the Court,” the order states:

Counsel are expected to abide by existing and evolving California State Bar guidance and advisory opinions on the use of AI in the legal profession, and counsel should conduct themselves in a manner consistent with ABA Resolution 604’s (Feb. 6, 2023) admonishment regarding accountability of individuals and organizations for any use of AI products, systems, and capabilities (and the Resolution’s provisions regarding operators of AI systems and capabilities). See https://perma.cc/A6WC-6X6P. As with any prudent approach in an evolving area of law, counsel and pro se parties are expected to be competent and knowledgeable about evolving judicial and legal standards in the use of generative AI tools, including case law and opinions addressing such issues (not limited to case law imposing sanctions for failure to use generative AI in an ethical manner).

In another part of §VII.C, titled “AI and Evidence,” the Standing Order states:

If a Party or counsel chooses to use an AI or other automated tool as part of a process for creating exhibits, demonstratives, or other material to be filed or presented to the Court, they shall only do so consistent with their ethical and legal obligations and shall use such tools responsibly and with competent training, knowledge, and understanding of the limitations and risks of such automated tools.

The court held a Discovery Management Conference (“DMC”) and entered an order consistent with that conference. It wrote: “These rulings are WITHOUT PREJUDICE as to requesting potential modifications once the Parties reach Phase Two of discovery, if sufficient good cause is shown to warrant modifying the Protective Order at that time.” 2026 WL 1483493, at *4.

What did the defendant want? “With regard to Section 7.1, Defendants proposed changes from the Model Protective Order primarily addressing the use of Confidential/Highly Confidential Information with an Artificial Intelligence (AI) tool…. 

The Litton court ordered that “Section 7.1 SHALL not be edited as proposed and the Model Protective Order language shall be used for the Protective Order in these cases. Further, the Court ORDERS that Section VII(C) of the undersigned’s Civil Standing Order regarding artificial intelligence is now applicable to these cases….”

The supporting analysis was:

At the DMC, the Parties and the Court noted that the undersigned’s Standing Order for Civil Cases has multiple provisions addressing Artificial Intelligence tools and the responsible use of such tools by lawyers (and experts and Parties generally). See Standing Order for Civil Cases at Section VII(C). The Court ordered at the DMC and hereby confirms that it is ORDERED that Section VII(C) of the undersigned’s Standing Order for Civil Cases shall apply to all matters within the scope of referral to the undersigned in these cases, including these discovery matters. Because there already exists a set of provisions addressing AI tools in that Standing Order, Defendants did not demonstrate sufficient cause to overcome the presumption of reasonableness and to warrant editing the Model Protective Order to address those same issues.

Id. at *4 (emphasis added).

I do not practice in the Northern District of California; however, it appears to me that the Standing Order focuses on materials to be filed in court. It addresses, for example, “in the course of preparing filings with the Court or other documents for submission in an action,” “AI and Briefs/Pleadings Filed with the Court,” and “a process for creating exhibits, demonstratives, or other material to be filed or presented to the Court….” 

It seems to me, however, that Defendants’ request was focused on the use of discovery materials outside of court. Perhaps, that may be deemed to be “in the course of preparing filings with the Court,” however, if so, it is not entirely clear to me.

A.I. protective or confidentiality orders are becoming a standard of care.

Michael D. Berman, Owner, E-Discovery LLC.

While the court’s stipulated protective order is presumptively reasonable, times change. I have suggested that A.I. protective or confidentiality orders are becoming a standard of care.

It is unclear to me whether Defendants in Litton were requesting limits on an opponent’s right to upload materials produced by Defendants in discovery to Plaintiffs’ A.I. tools. Nor is it clear whether it was directed at or limited to consumer-grade or secure A.I. tools.

To the extent, if any, to which Defendants were seeking a protective or confidentiality order prohibiting the Plaintiffs from uploading confidential and highly confidential materials to A.I. tools that do not meet security standards, it seems that they were likely entitled to relief.

Further, uploads of an opponent’s discovery production to A.I. may impact clawback and privilege non-waiver agreements. Assume that parties negotiate a Fed.R.Evid. 502 order and Fed.R.Civ.P. 26(b)(5) “clawback.” Assume no limitation on a receiving opponent’s use of A.I., when discovery materials are produced to it. If the producing party inadvertently produced privileged material that was uploaded to an opponent’s A.I., how does the recipient terminate use and “return, sequester, or destroy” it when the “clawback” is asserted? 

Uploads of an opponent’s discovery production to A.I. may impact clawback and privilege non-waiver agreements. 

Michael D. Berman, Owner, E-Discovery LLC.

One may debate whether such a provision is now standard of care for protective orders. However, it is clear that such provisions are being utilized. See A.I. Protective Orders Are Becoming Routine.

It is interesting to note a related recent decision in a criminal case. U.S. v. Hoskins, 2026 WL 1493698 (N.D. Ill. May 28, 2026).

There, on an agreed motion of the government, the court entered an “Amended Protective Order Governing Discovery” under Fed.R.Crim. P. 16(d) and 18 U.S.C. §3771(a)(1) and (8). Paragraph 6 states:

Restrictions on Use of AI to Process Materials. No person or entity authorized to have access to Materials under the terms of this Order shall input, transmit, upload, process, generate output from, or otherwise expose any Materials received pursuant to this order to any artificial intelligence (“AI”) tool without prior written notice to, and corresponding receipt of acknowledgement from, the government. “AI tool” means any automated system that uses statistical modeling, machine learning, or similar techniques to process inputs and generate outputs, recommendations, or predictions, including but not limited to large language models, generative AI services, or AI-assisted software tools, whether cloud-based or otherwise. When providing written notice for the otherwise prohibited use of an AI tool, defense counsel must specifically identify and describe any AI tool to be used and certify that it does not (i) retain or use any Materials for model training, nor (ii) expose any Materials to third parties not authorized under this order to receive such materials. In addition, defense counsel must certify that: (i) defense counsel has implemented reasonable measures to ensure that any AI tool used will maintain the confidentiality of any Materials, and (ii) defense counsel will ensure the deletion of all Materials from the tool at the conclusion of this case. Defense counsel will be responsible for destroying such information at that time. Under no circumstances will any member of the defense team submit Materials to a publicly accessible AI system that retains and uses submitted data to train models. Such AI tools pose unique risks to the security and integrity of Materials, given the practical inability to claw back or delete data once it has been incorporated into a model.

Id. at *2 (emphasis added).

Assisted by GAI and LLM Technologies per EDRM’s GAI and LLM Policy.

Author

  • Michael Berman

    Michael Berman is a practicing lawyer and an adjunct faculty member at the University of Baltimore School of Law.  He has published extensively, including as the editor-in-chief and a contributing author in “Electronically Stored Information in Maryland Courts” (Maryland State Bar Ass’n. 2020), co-editing two American Bar Association books on electronic discovery, as well as co-authoring law review and other articles regarding electronically stored information (“ESI”).  He has presented widely in venues ranging from local to National events and served as a Court Appointed ESI Discovery Supervisor for ESI Protocol.

    View all posts Owner, E-Discovery LLC