Air-Evac EMS, Inc. et al v. Aetna Life Insurance Company et al — Court restricts use of confidential discovery material in generative AI tools

Background

Air-Evac EMS, Inc., Med-Trans Corporation, Guardian Flight LLC, and Reach Air Medical Services LLC are the plaintiffs, and Aetna Life Insurance Company and Aetna Health, Inc. are the defendants. The November 4, 2026 order was entered in the United States District Court for the District of Connecticut in Civil Action No. 3:26-cv-01021-SVN. The defendants sought an unopposed agreed supplement to the standing protective order previously entered at docket entry 9, and the plaintiffs did not oppose the motion.

Acting under Federal Rule of Civil Procedure 26(c) and HIPAA, the court found good cause to modify the existing order. The supplement expands the definition of confidential information to include proprietary technical, scientific, financial, business, health, and medical information, as well as protected health information and confidential health information. It also addresses disclosure to prior authors and recipients, adjudicators, and former employees; the timing of confidentiality designations; inadvertent production or disclosure; retention of litigation files; and protections for nonparties. The order does not decide the merits of the parties’ underlying claims.

The Court’s Holding

The court entered the agreed supplement and added paragraph 24, titled “Use of Discovery Material in Conjunction with Generative Artificial Intelligence.” Designated Material may not be input into generative AI tools or platforms, including tools substantially similar to the examples listed, except through two channels: secure electronic discovery platforms whose access is limited to attorneys working on the action, and enterprise generative AI tools that do not train on or retain submitted data and do not disclose data to third parties. The order identifies Harvey and Legora as examples of the latter category. Providing Designated Material to a tool outside these exceptions is treated as disclosure to a third party.

Outputs from a permissible generative AI tool are deemed attorney work product when they contain, reflect, incorporate, or derive from Designated Material. The same paragraph extends work-product protection to prompts, queries, instructions, inputs, outputs, logs, model or tool settings, and similar materials created for the litigation, including materials connected with expert work. That protection does not eliminate disclosures required by Rule 26(a)(2), and the court retains authority to order disclosure for good cause.

The supplement separately authorizes the parties to receive, subpoena, transmit, or disclose PHI concerning the medical claims at issue, subject to the order and applicable law. It excludes specified categories of nonparty information from party-to-party receipt, including information concerning HIV or AIDS, mental or emotional conditions or disorders, and genetic testing, while allowing further orders where necessary to comply with state requirements. Confidentiality and use restrictions continue after the litigation, disputes may be brought to the court by motion, and violations may result in sanctions and costs.

Key Takeaways

• The order prohibits placing Designated Material into generative AI platforms unless the material stays within a secure, attorney-only e-discovery platform or a qualifying enterprise tool.
• A qualifying enterprise tool must not train on or retain submitted data and must not disclose data to third parties.
• Using an AI platform outside those exceptions is expressly treated as a third-party disclosure of protected discovery material.
• AI-generated outputs derived from Designated Material, along with prompts, logs, settings, and related litigation materials, receive attorney-work-product treatment subject to Rule 26(a)(2) and good-cause disclosure orders.
• The order governs discovery handling and AI use rather than resolving the underlying dispute between the air-medical providers and Aetna entities.

Why It Matters

The decision provides a concrete example of a federal court addressing generative AI through discovery governance rather than through sanctions or a ruling on AI-generated advocacy. Its central distinction is architectural and operational: access-limited e-discovery systems and enterprise tools with specified data-handling restrictions are permitted, while other platforms are treated as third-party recipients. The order also connects AI use to work-product doctrine without making that protection absolute, preserving expert-disclosure obligations and judicial oversight.

The supplement’s broader confidentiality provisions matter to the AI restriction because the protected material includes commercially sensitive information and health data. The order permits litigation-related PHI handling under stated safeguards but excludes particular nonparty medical categories and preserves independent confidentiality duties. Its continuing obligations, nonparty protections, and enforcement provisions make the AI rules part of a comprehensive discovery framework rather than a standalone technology policy.