Colorado Court Bars Confidential Discovery Data From Most Free AI Tools

A discrimination case becomes an AI privacy dispute

Morgan v. V2X, Inc. began as an employment discrimination case. It turned into a dispute over how confidential discovery materials are handled when AI is involved. Everlaw, which analyzed the decision, says it is the first order of its kind as far as the company is aware.

The plaintiff represents himself and wanted to use AI tools to narrow the technological gap with a well-funded corporate defendant. The defendant, V2X, Inc., worried that confidential information, including trade secrets and personnel files, was being fed into mainstream platforms such as ChatGPT or Gemini. Everlaw explains the concern this way: if models are trained on confidential information, they may replicate some or all of it in future outputs.

The plaintiff argued that his choice of software was protected work product. Magistrate Judge Maritza Dominguez Braswell disagreed. She ruled that a party's mental impressions are protected, but the identity of the AI tool is not. If confidential data goes to a platform that may compromise its confidentiality, the opposing side has a legitimate right to know where it is going.

What the court said about privacy and chatbots

The court discussed the Fourth Amendment and reasonable expectations of privacy. It acknowledged that in 2026 almost all data passes through third-party systems such as Gmail, and asked whether anyone with a Gmail account has forfeited all rights to confidentiality and privacy. It answered no. The order said the Fourth Amendment offers "a wholly different legal framework from the work product doctrine," but that the principle is informative: "routing information through a third-party system does not forfeit all privacy."

Judge Braswell also drew a line between search engines and chatbots. She wrote that many modern AI platforms are "specifically designed and trained to engage," invite "candid and significant disclosure of information," and simulate empathy and foster trust. According to the order, research confirms people share sensitive information with chatbots, often without appreciating what happens to it afterwards.

The court also said such interactions by a pro se litigant closely resemble the work product that Rule 26(b)(3) was designed to protect. Everlaw notes the court left that point at "may." The plaintiff still lost on this issue, because he had not carried his burden to show that disclosing the tool would reveal mental impressions or strategy here. The court concluded that routing data through an AI system does not automatically waive all protections, but it creates a risk that mainstream, low-cost tools will persistently collect and store data for their own purposes.

The three contractual requirements

The modified protective order bars any party or authorized recipient from inputting, uploading or submitting CONFIDENTIAL information into any AI platform unless the provider is contractually prohibited from two things: storing or using inputs to train or improve its model, and disclosing inputs to any third party except where essential to delivering the service. Any such third party must be bound by obligations no less protective than those in the order.

Everlaw summarizes the rules as three requirements:

• Training bans. The provider must be contractually prohibited from using inputs to train or improve its models.
• Strict confidentiality. The provider cannot disclose inputs to third parties unless essential for service delivery.
• Right to delete. The user must have the contractual ability to remove or delete all confidential information on request.

A party that believes its AI tool meets these terms must retain written documentation of the contractual protections.

Impact on pro se litigants and practitioners

The court acknowledged the cost of its approach. It wrote that the provision will, "at least for now," bar the parties from using most, if not all, mainstream low-to-no cost AI to process Confidential Information, and that this "type of restriction disadvantages pro se litigants." The decision also noted that enterprise-tier AI accounts are often available only through organization-wide procurement and are too expensive for pro se litigants.

The court stressed the limits of the order. It said it does not intend to leave the plaintiff without the benefits of AI, because modern AI tools may be used in many ways that do not involve uploading Confidential Information, and nothing in the order restricts those uses.

For practitioners, Everlaw's reading is that the decision effectively bars standard or free versions of mainstream AI for processing sensitive case data. The burden, it says, is now on the attorney to verify the terms of service. Everlaw frames the stakes as ensuring a client's data does not become "the training set for tomorrow's algorithm."