Part 3a of What is the role of Gen AI in Litigation Series

How AI is impacting the Court’s approach to Confidential Information

AI creates new risks for preserving confidentiality - its use by parties involved in litigation raises important questions about protecting confidential information in the context of commercial disputes. Trade secrets and commercially sensitive information are often disclosable in court proceedings, but protective orders may be put in place to prevent this material from reaching competitors or entering the public domain.

When sensitive materials are input into AI platforms, they may be stored, used for training large language models, or shared with third parties. Confidentiality agreements and protective orders that predate the availability and growing use of Gen AI may therefore fail to adequately address these risks.

In Morgan v V2X, Inc., the US District Court for the District of Colorado addressed these issues directly.

Can the court order disclosure of the AI tools being used?

In Morgan, the defendant sought an order compelling the claimant to disclose the identity of the AI tool he was using, so it could determine whether the tool had appropriate safeguards in place to protect its confidential information. The claimant resisted, arguing that the “specific software, research platforms, and analytical tools chosen by a litigant to review discovery, synthesize information, and prepare for trial fall squarely under the Work-Product Doctrine, codified in Federal Rule of Civil Procedure 26(b)(3)”.

The court found that the defendant’s request was legitimate and reasonable. Whilst the use of an AI tool does not automatically waive work-product protection (see AI in litigation: Texas Business Court addresses work product protection and discovery disclosure requirements) in this case the claimant had not demonstrated that disclosing the identity of the AI tool would reveal his mental impressions or strategy, and so the information sought was not protected by the Work Product Doctrine. As such, the court ordered the claimant to disclose the name of any AI platform used to upload, submit, process, review, analyse, organise, or store any information designated as confidential.

Can the court restrict the use of AI tools by litigants?

In addition to seeking disclosure of the AI tool being used, the defendant asked the court to amend the existing protective order with new language to address the claimant’s use of AI. The court did not accept the defendant’s proposed wording, but it did agree to amendments restricting AI use. The practical effect was significant:

“[N]o party or authorised recipient may input, upload or submit [CONFIDENTIAL] Information into any modern AI platform, including any generative, analytical, or large language model based tool (“AI”), unless the AI provider is contractually prohibited from (1) storing or using inputs to train or improve its model, and (2) disclosing inputs to any third party except where [such disclosure is] essential to facilitating delivery of the service. Where disclosure to a third party is essential to service delivery, any such third party shall be bound by obligations no less protective than those required by this Order. In addition, the AI provider must contractually afford the party or authorized recipient the ability to remove or delete all [CONFIDENTIAL] information upon request. A party intending to use AI that it contends meets these requirements must retain written documentation of these contractual protections”.

The court recognised that this restriction would, in practice, bar the parties from using most mainstream low or no cost AI to process “Confidential Information” (as defined in the proceedings). Accordingly, it cautioned the parties against “the over-designation of Confidential Information” and indicated that the scope of what qualifies as “confidential” would have to be narrowly construed.

We are yet to see how the English courts might address the same issue and the extent to which they might be prepared to restrict a parties use of certain AI tools.

Do your existing confidentiality orders or agreements still offer protection?

In these proceedings the current Protective Order arguably covered the use of AI already because the Order broadly prohibited disclosure, not just to a person or entity, but generally and in any fashion, except under specifically enumerated circumstances.  Therefore disclosure in the form of a transmission to an AI system with a provider that stores the confidential information in their own databases, and/or for their own purposes, could in itself have been a violation of the current Protective Order.  In this case, however, the parties agreed that a clarifying amendment was necessary.

Any existing confidentiality agreements or protective orders that do not expressly address AI should be reviewed to assess whether they offer sufficient protection. If an amendment is necessary, consider seeking a variation. The Morgan decision provides helpful guidance on how courts might approach similar applications. The court in Tate Group Automotive, LLC v. Legacy Automotive Capital, LLC recommended that the parties amend their protective order to make clear whether, how and to what extent confidential information may be shared with an AI tool or other large language model system.

New applications to restrict AI use for reasons of confidentiality

Where no protective order is yet in place, we may see a rise in parties seeking orders that expressly address AI and restrict its use. The Morgan decision provides useful guidance on the factors likely to influence judicial discretion. Restrictions on AI use for reasons of confidentiality may, however, be narrow in scope. Courts will scrutinise overly broad designations of confidential information, particularly where they disproportionately burden unrepresented litigants who cannot access enterprise AI tools.

At what point does ‘confidential information’ cease to be confidential?

One of the common questions that arises in relation to the use of AI is whether uploading information to an AI tool, particularly publicly available open source tools such as Chat GPT, Gemini or Claude, results in the loss of its confidentiality.

Some US courts have taken the approach that, absent a restrictive agreement or order to the contrary, uploading confidential information to public AI tools may not constitute a work-product waiver by a self-represented party because it is not a disclosure to “an adversary or in a way likely to get in an adversary's hand”. (See What is the role of Gen AI In Litigation? Part 2: AI, Legal Advice and Privilege)

The question of whether a non-lawyer’s AI-related communications are protected from discovery, and what happens when documents produced in discovery are shared with an AI tool is addressed in our briefing here: AI in litigation: Texas Business Court addresses work product protection and discovery disclosure requirements.

In the UK, confidentiality can be lost as a result of disclosure to a third party (without appropriate privilege protections or agreements as to confidentiality in place) and/or by information entering the “public domain”.

The recent Court of Appeal decision in Illiquidx Limited v Altana Wealth and others [2026] EWCA Civ 874 has offered some timely guidance on what is meant by the words “in the public domain”, confirming that it is information which is "so generally accessible that, in all the circumstances, it cannot be regarded as confidential". In that case the appellants argued for a broader meaning of entering the "public domain" to cover any information which had been disclosed to anyone without being subject to a duty of confidentiality, even if it was not generally accessible.  The Court of Appeal rejected this in re-stating that “the true criterion [of confidentiality] is not secrecy… but inaccessibility” and holding that “Relative inaccessibility can suffice to make information confidential rather than in the public domain”. It should though be noted that the Court recognised that test of whether information has entered the “public domain” is a fact-sensitive one with potential grey areas.

It will be interesting to see how the concept of “entering the public domain” may continue to develop in the UK when it comes to uploading confidential information to publicly accessible AI tools. Will it be an outright loss of confidentiality as suggested by the Upper Tribunal Immigration and Asylum Chamber, or might the position going forward be more nuanced?

Key takeaways

1. Courts may permit requests to disclose the identity of AI platforms being used in litigation, particularly where there is reason to believe confidential information has been uploaded to an AI system.
2. Opposing parties may seek disclosure of the identity of the AI platforms used and the contractual safeguards applicable to them. Retaining a record of the contractual protections in place may assist in responding to these requests.
3. Review whether existing protective orders address AI use. If not, consider whether to seek variations to include AI specific restrictions, particularly where sensitive commercial information or trade secrets are at stake.
4. Courts are likely to require that the scope of confidential information be narrowly construed. In Morgan the court cautioned parties against the “over-designation of Confidential Information”. Where AI restrictions would significantly burden a party’s ability to conduct litigation, courts may insist on a more limited definition of what qualifies as confidential. Parties seeking to restrict AI use should be prepared to justify their confidentiality designations.