Practitioners should proactively propose AI-specific protections—prefer Model 3—and identify compliant AI tools before litigation, while never submitting privileged information to public AI platforms and retaining documentation of contractual protections.

Federal Rule of Civil Procedure 26(c)(1)(G) provides the basis for AI-specific protective-order restrictions, and courts treat open AI data submissions as potential waivers of attorney-client privilege.

There are four emerging protective-order models: Blanket Prohibition, Data-Rights Approach, Contractual Safeguards Approach, and Notice-and-Secure-Environment Approach, with Morgan v. V2X and Jeffries v. Harcros guiding influence.

Courts are rapidly adding protective-order language to limit the use of generative AI with confidential discovery materials due to concerns about data retention, training, and third-party disclosure.

The Sixth Circuit in United States v. Farris underscores ethical obligations: attorneys must use AI consistent with ethics rules to protect client confidentiality and privilege, with sanctions for incompetence or candor violations.

The privilege vs. work product distinction matters: inputs to AI platforms can waive attorney-client privilege as third-party submissions, while work product protections may remain depending on context and case, as seen in Warner v. Gilbarco and Morgan v. V2X.

Action steps for current practice include implementing contractual safeguards, maintaining written records of AI protections, educating teams and clients about AI risks, and monitoring ongoing case law and guidance.