Two Federal Courts. One Week. Opposite Answers. And Missouri Practitioners Are Flying Blind.

Kirksville Attorney Patrick Nolan

Nolan Law Firm LLC

Disclaimer: This analysis is provided for informational and educational purposes only and does not constitute legal advice. The opinions expressed are those of the author. Readers should consult qualified counsel for advice on specific situations.

I. The Question Nobody in Missouri Has Answered

Here is the problem: more than half of U.S. households have adopted AI in some form. Lawyers are using it. Clients are using it. Pro se litigants are using it. And in Missouri, no appellate court has told anyone what happens to the information they feed into these platforms when the other side asks for it in discovery.

If you are a Missouri attorney using a consumer-tier AI platform to analyze case strategy, you may be waiving privilege on everything you type into it. If you are opposing a pro se litigant who built their case with ChatGPT, you may or may not be able to discover that process. Nobody knows, because no Missouri court has ruled on the question.

In February 2026, two federal courts tried to answer it. They reached opposite conclusions. Those opinions, issued one week apart, now frame the national debate. But they are both district-level rulings with no binding authority beyond their own cases, and neither one applies Missouri law.

This analysis examines both opinions, maps their reasoning onto Missouri’s existing privilege framework, and identifies the practical questions that Missouri practitioners need to resolve before, not after, they find themselves arguing about AI discoverability in a Kirksville or Kansas City courtroom.

II. The Federal Split: Warner and Heppner

A. Warner v. Gilbarco: AI Is Just a Tool

On February 10, 2026, Magistrate Judge Anthony Patti in the Eastern District of Michigan issued an order in Warner v. Gilbarco, Inc., Case No. 2:24-cv-12333, denying the defendants’ motion to compel a pro se plaintiff’s AI usage in litigation. The defendants wanted everything: the plaintiff’s AI inputs, outputs, chat logs, and any analysis generated through generative AI platforms. They argued that using ChatGPT to prepare litigation materials constituted a waiver of work product protection.

Judge Patti rejected every part of that request. His reasoning rested on three pillars. First, the materials were prepared in anticipation of litigation and fell squarely within the work product doctrine under Fed. R. Civ. P. 26(b)(3)(A). Second, the plaintiff, as a pro se litigant, retained the right to assert work product protection over her litigation preparation materials. Third, and most significantly, Judge Patti rejected the waiver argument by holding that “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.”

The court also addressed proportionality. Even if the AI usage materials were marginally relevant, the request was disproportional under Rule 26(b)(1). In a pointed footnote, Judge Patti stated: “Defendants’ preoccupation with Plaintiff’s use of AI needs to abate.”

The court’s closing observation framed the dispute symmetrically: “In the end, both sides of this dispute seek to obtain each other’s thought processes, while shielding their opponent from discovery of their own. The Court will uphold the protections afforded the thought processes and litigation strategies of both sides and will order production of neither.”

B. United States v. Heppner: The Corporate Entity Behind the Tool

Seven days later, on February 17, 2026, Judge Jed Rakoff in the Southern District of New York issued a memorandum opinion in United States v. Heppner, Case No. 25-cr-503, reaching the opposite conclusion. The government moved for a ruling that approximately thirty-one documents memorializing the defendant’s communications with Claude (the AI platform operated by Anthropic) were protected by neither attorney-client privilege nor the work product doctrine.

Judge Rakoff granted the motion in its entirety. His analysis was methodical and fact-specific, and it demolished the privilege claim on all three elements of the attorney-client privilege test.

First, the communications were not between the defendant and his attorney. Claude is not an attorney, and “in the absence of an attorney-client relationship, the discussion of legal issues between two non-attorneys is not protected by attorney-client privilege.” The court dismissed the argument that AI inputs are analogous to using cloud-based word processing, noting that the use of such applications “is not intrinsically privileged in any case” and that all recognized privileges require “a trusting human relationship” with “a licensed professional who owes fiduciary duties and is subject to discipline.”

Second, the communications were not confidential. This is where Rakoff went somewhere Judge Patti never went. He examined Anthropic’s actual privacy policy and found that users consent to Anthropic collecting data on both inputs and outputs, using that data to train the AI, and reserving the right to disclose it to “governmental regulatory authorities.” The court concluded that Heppner “could have had no reasonable expectation of confidentiality in his communications” with Claude.

Third, the defendant did not communicate with Claude for the purpose of obtaining legal advice. Claude itself disclaims providing legal advice. When the government asked Claude directly whether it could give legal advice, it responded: “I’m not a lawyer and can’t provide formal legal advice or recommendations.” The fact that Heppner later shared the outputs with counsel did not create retroactive privilege, because “non-privileged communications are not somehow alchemically changed into privileged ones upon being shared with counsel.”

On work product, Rakoff was equally direct. The AI Documents were not “prepared by or at the behest of counsel.” Defense counsel conceded that he “did not direct [Heppner] to run Claude searches.” Because the documents were prepared by the defendant on his own volition, without counsel’s direction, and did not reflect counsel’s strategy at the time of creation, they did not qualify for work product protection.

C. The Critical Distinction

Judge Patti treated AI as a tool, like a word processor or a legal research database. Under that framing, the person behind the keyboard is just doing legal work, and the method of production is irrelevant.

Judge Rakoff looked past the tool to the corporation operating it. Anthropic collects your inputs. It has employees who can access them. Its privacy policy reserves the right to share them with third parties, including the government. That is not a hammer. That is a third-party corporate relationship governed by a contract you agreed to when you clicked “Accept.”

Patti’s framework sounds reassuring, but it collapses when you examine it closely. If AI is just a tool and not a third party, then the privacy policy should not matter. The data retention practices should not matter. The terms of service should not matter. A wrench does not have a privacy policy. But the moment you acknowledge that any of those things are relevant, you have already conceded that there is a corporate entity on the other side of the transaction. And that entity is the third party.

Rakoff’s framework is the more analytically sound approach. The AI is the tool. The corporation is the third party. The question is what contractual and technical safeguards govern that third-party relationship.

III. Weight of Authority and Procedural Posture

Neither opinion binds anyone outside its own case. But they do not carry equal weight.

Judge Patti is a magistrate judge in the Eastern District of Michigan. His order is reviewable by the district judge under Fed. R. Civ. P. 72(a) and could be overturned within the same case on a standard of “clearly erroneous or contrary to law.” Judge Rakoff is a district court judge in the Southern District of New York, one of the most influential trial courts in the country, with a reputation for issuing opinions that shape national legal discourse. His memorandum opinion styled itself as addressing “a question of first impression nationwide.”

Appellate courts have not addressed the question in any circuit. Until they do, both opinions remain persuasive authority only. But if you are a practitioner trying to decide which framework to follow in the absence of binding precedent, the safe bet is Rakoff. If you are wrong about Patti’s approach, you have waived privilege on everything your client fed into a consumer AI platform. If you are wrong about Rakoff’s approach, you have been overly cautious. One of those mistakes is recoverable. The other is not.

For Missouri practitioners specifically, neither opinion applies. Missouri state courts apply Missouri privilege law in state proceedings, and the Missouri Supreme Court and Court of Appeals have their own body of case law on confidentiality, third-party disclosure, and waiver. The question that matters is which direction Missouri courts will go when the issue arrives.

IV. Missouri’s Existing Privilege Framework

No Missouri Supreme Court or appellate court decision has directly addressed whether case information submitted to an AI platform is discoverable. But Missouri’s existing authority on attorney-client privilege, work product, and third-party disclosure provides a clear analytical framework. The key Missouri authorities track Rakoff’s approach far more closely than Patti’s.

A. Attorney-Client Privilege Under Missouri Law

Missouri’s attorney-client privilege is codified at Mo. Ann. Stat. § 491.060(3). Under State v. Longo, 789 S.W.2d 812 (Mo. App. E.D. 1990), the privilege requires four elements: (1) the communication was transmitted by voluntary act of disclosure; (2) it was between a client and a lawyer; (3) it was made in confidence; and (4) it was made by means that, so far as the client is aware, discloses the information to no third parties other than those reasonably necessary for the transmission of the information or for the accomplishment of the purpose for which it is transmitted.

All four elements must be present. When a party submits case information to a consumer AI platform operating under terms and conditions that grant the company independent rights to use that information, elements three and four are immediately at risk. The communication is not “in confidence” when the platform operator reserves the right to access, use, and disclose it. And the information is disclosed to a third party, the corporate entity operating the platform, that is not “reasonably necessary” for the attorney-client communication.

B. Third-Party Disclosure and Waiver

Missouri law is clear that voluntary disclosure to a third party waives attorney-client privilege. In State ex rel. Garrabrant v. Holden, 633 S.W.3d 356 (Mo. 2021), the Missouri Supreme Court held that waiver through third-party disclosure does not require subjective intent. Conduct that “would make it unfair for the client to invoke the privilege thereafter” suffices, because “the essential function of the privilege is to protect a confidence that, once revealed by any means, leaves the privilege with no legitimate function to perform.”

This is directly applicable to AI platform usage. Submitting case information to a commercial AI platform is a voluntary act. The user chose to type the information and press enter. The only question is whether that submission constitutes disclosure to an “independent third party” or remains within a protected confidential relationship.

Missouri courts have long held that the presence of a third party not essential to the transmission of information vitiates the privilege. Kratzer v. Kratzer, 595 S.W.2d 453 (Mo. App. E.D. 1980). When the AI platform’s terms of service grant the company independent rights to the submitted data, including training rights, quality control access, and potential disclosure to regulatory authorities, the company is not acting as a confidential agent. It is acting as an independent commercial entity with its own interests.

C. The Confidential Agent Exception

Missouri recognizes that privilege is not destroyed when the third party acts as a confidential agent of either the client or the attorney. McCaffrey v. Brennan’s Estate, 533 S.W.2d 264 (Mo. App. 1976). Under McCaffrey, “it is the function of the third party, his relationship to the client, which determines whether the privilege encompasses communications made by, or in the presence of that third party.”

This exception is the strongest argument for preserving privilege when using AI platforms. If the AI tool functions as a confidential agent, analogous to a consulting expert or litigation support vendor, the privilege might survive. But the exception depends on the nature of the relationship. A confidential agent acts for the benefit of and under the control of the principal. A company that retains independent data rights operates for its own benefit and outside the attorney-client relationship. Terms of service that grant the company independent rights to use, train on, or exploit submitted data signal a relationship fundamentally at odds with that of a confidential agent.

This is where the distinction between consumer-tier and enterprise-tier AI plans becomes legally significant. A consumer free plan with broad data usage rights in the terms of service looks nothing like a confidential agent relationship. An enterprise agreement with contractual prohibitions on data retention and use, a data processing addendum, and explicit confidentiality obligations looks much more like one.

D. Work Product Under Missouri Law

Missouri’s work product doctrine, codified in Rule 56.01(b)(5), provides qualified protection for documents and tangible things prepared in anticipation of litigation by or for a party or its representative. The Missouri Supreme Court in State ex rel. Ford Motor Co. v. Westbrooke, 151 S.W.3d 364 (Mo. 2004), held that the doctrine protects both tangible work product (written statements, briefs, memoranda) and intangible work product (mental impressions, conclusions, opinions, legal theories). Intangible work product receives absolute protection and cannot be discovered even upon a showing of substantial need. State ex rel. Atchison, Topeka & Santa Fe Ry. Co. v. O’Malley, 898 S.W.2d 550 (Mo. 1995).

For work product, Missouri follows a more flexible waiver analysis than for attorney-client privilege. Under O’Haver v. 3M Co., 698 S.W.3d 730 (Mo. App. W.D. 2024), applying Edwards v. Mo. State Bd. of Chiropractic Examiners, 85 S.W.3d 10, 27 (Mo. App. W.D. 2002), “a disclosure made in the pursuit of trial preparation and not inconsistent with maintaining secrecy against opponents” does not waive work product protection.

This is the strongest doctrinal basis for protecting AI-assisted litigation work under Missouri law. If a lawyer submits case materials to an AI platform for analysis as part of trial preparation, and the opposing party has no access to the platform’s analysis or the submitted data, the disclosure might satisfy the O’Haver standard. The AI platform is being used in pursuit of trial preparation, and the opposing party does not have access to the results.

But there is a critical weakness. The O’Haver standard requires that the disclosure be “not inconsistent with maintaining secrecy against opponents.” When the AI platform’s terms of service grant the company independent rights to use, store, and potentially disclose the submitted data, the information is available to a commercial enterprise for its own purposes. That availability, even if the opposing party does not currently have access, may be “inconsistent with maintaining secrecy” in a functional sense. The work product doctrine protects the adversarial process, and sharing litigation strategy with a corporation that has no obligation to keep it secret from anyone undermines that purpose.

V. The Practical Line: Contract Terms as the Determinative Factor

Both the federal case law and the Missouri framework point to the same practical conclusion: the contractual relationship between the user and the AI platform operator determines whether privilege survives.

Under a consumer free-tier plan with standard terms of service, the analysis runs strongly against privilege. The platform operator collects inputs and outputs. It may use the data for model training. It reserves the right to disclose to third parties. There is no confidentiality obligation running from the operator to the user. Under Rakoff’s framework, there is no reasonable expectation of confidentiality. Under Missouri’s Garrabrant and Kratzer line of cases, there is a voluntary disclosure to an independent third party that destroys confidentiality.

Under an enterprise or API plan with a negotiated data processing agreement, the analysis can shift. If the agreement prohibits the operator from retaining or using submitted data for any purpose other than providing the requested service, if it includes explicit confidentiality obligations, and if it contractually prohibits disclosure to third parties absent legal compulsion, the relationship starts to look more like a confidential agent arrangement under McCaffrey. The user has taken “reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of the client,” as required by Missouri Rule of Professional Conduct 4-1.6(c).

The distinction is not theoretical. It is the difference between a lawyer who dictates notes to a secretary bound by a confidentiality agreement and a lawyer who reads those same notes aloud in a crowded coffee shop. The tool is the same in both cases; the relationship and the environment are not.

VI. Missouri Rule 4-1.6 and Professional Obligations

Beyond the discovery context, Missouri Rule of Professional Conduct 4-1.6 imposes independent confidentiality obligations on attorneys. Rule 4-1.6(a) prohibits revealing “information relating to the representation of a client” without informed consent, implied authorization, or an enumerated exception. Rule 4-1.6(c) requires “reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of the client.”

Agreeing to an AI platform’s terms of service that grant the company rights to access and use submitted case information may itself constitute a disclosure, or at minimum a failure to take reasonable precautions, in violation of Rule 4-1.6. The Comments to Rule 4-1.6 explain that the confidentiality rule applies not only to matters communicated in confidence by the client but also to “all information relating to the representation, whatever its source,” and extends to “disclosures by a lawyer that do not in themselves reveal protected information but could reasonably lead to the discovery of such information by a third person.”

While Rule 4-1.6 does not itself determine discoverability, a court analyzing whether privilege was waived by AI disclosure may look to whether the attorney complied with professional conduct obligations as evidence of whether the disclosure was intended to remain confidential.

VII. Missouri Circuit Court AI Disclosure Rules

Multiple Missouri circuit courts have enacted local rules requiring disclosure when AI tools are used to prepare filings. The Seventh Judicial Circuit (Clay County), the Seventeenth Judicial Circuit (Cass and Johnson Counties), the Twentieth Judicial Circuit (Franklin, Gasconade, and Osage Counties), the Fourth Judicial Circuit, and the Forty-Sixth Judicial Circuit (Taney County) all require litigants and attorneys to disclose the specific AI tool used when generative AI is employed to conduct legal research or draft pleadings.

These rules govern disclosure to the court, not the discoverability of case information submitted to AI platforms. But they reflect a growing judicial awareness of AI use in litigation and suggest that Missouri courts are paying attention. The absence of any statewide Missouri Supreme Court rule or bar ethics opinion specifically addressing AI confidentiality and privilege means that the analysis currently rests on applying general privilege principles to a new technology context.

VIII. Practical Guidance for Missouri Practitioners

Until Missouri appellate courts address this question, practitioners should operate under the assumption that Rakoff’s framework, not Patti’s, will prevail. That means treating the platform operator as a third party and structuring your AI usage accordingly.

First, review the terms of service before submitting any privileged or work-product-protected materials to an AI platform. Platforms whose terms grant the company rights to use, train on, or otherwise exploit submitted data present a material risk of privilege waiver under existing Missouri doctrine.

Second, use AI tools that operate under confidential processing agreements, data processing addendums, or enterprise agreements that explicitly prohibit the operator from retaining or using submitted data for any purpose other than providing the requested service. The contractual relationship is what will determine the privilege outcome.

Third, consider obtaining client informed consent before submitting case-specific information to any AI platform, and document that consent. If the question of waiver ever arises, evidence that the client was informed and consented to the specific platform’s data practices strengthens the argument that confidentiality was intended and maintained.

Fourth, segregate your AI usage. Use consumer-tier AI for general legal research, drafting templates, and non-privileged work. Use enterprise-tier or self-hosted AI for anything touching client-specific case strategy, privileged communications, or work product. The platform tier is not a technicality; under the emerging legal framework, it is the line between protected and discoverable.

Fifth, if you are on the other side and want to discover an opposing party’s AI usage, understand the limits. Both Warner and the Missouri work product doctrine create significant barriers to compelling disclosure of a litigant’s drafting process and thought patterns. The stronger discovery argument targets not the AI usage itself but potential violations of protective orders (uploading confidential documents to unprotected platforms) or specific factual assertions that may have been generated or influenced by AI analysis.

IX. Conclusion

Missouri practitioners are operating in a gap. No state appellate court anywhere in the country has ruled on whether AI-assisted litigation preparation is discoverable. The two federal district courts that have addressed the question reached opposite conclusions based on different analytical frameworks.

Missouri’s existing privilege law, particularly the Garrabrant waiver standard, the Kratzer third-party disclosure rule, the McCaffrey confidential agent exception, and the O’Haver work product analysis, provides the analytical tools to resolve this question. But until a Missouri appellate court applies those tools to the specific facts of AI platform usage, the answer remains uncertain.

The safest course is to treat AI platform operators as what they are: corporate third parties whose relationship to the attorney and client is governed by contract. If the contract protects confidentiality, the privilege argument is strong. If the contract reserves broad data usage rights, the privilege argument is weak. The tool is irrelevant. The corporate relationship is everything.

Whoever brings this question to the Missouri Court of Appeals first will shape the framework for every practitioner in the state. That case is coming. The only question is whether you will be prepared for it.

Sources and Authorities Cited

Federal Cases

Warner v. Gilbarco, Inc., Case No. 2:24-cv-12333 (E.D. Mich. Feb. 10, 2026) (Patti, M.J.)

United States v. Heppner, Case No. 25 Cr. 503 (S.D.N.Y. Feb. 17, 2026) (Rakoff, J.)

Missouri Cases

State v. Longo, 789 S.W.2d 812 (Mo. App. E.D. 1990)

State ex rel. Garrabrant v. Holden, 633 S.W.3d 356 (Mo. 2021)

Kratzer v. Kratzer, 595 S.W.2d 453 (Mo. App. E.D. 1980)

McCaffrey v. Brennan’s Estate, 533 S.W.2d 264 (Mo. App. 1976)

State ex rel. Ford Motor Co. v. Westbrooke, 151 S.W.3d 364 (Mo. 2004)

State ex rel. Atchison, Topeka & Santa Fe Ry. Co. v. O’Malley, 898 S.W.2d 550 (Mo. 1995)

O’Haver v. 3M Co., 698 S.W.3d 730 (Mo. App. W.D. 2024)

Edwards v. Mo. State Bd. of Chiropractic Examiners, 85 S.W.3d 10 (Mo. App. W.D. 2002)

State ex rel. Kilroy Was Here, LLC v. Moriarty, 633 S.W.3d 406 (Mo. App. E.D. 2021)

State ex rel. Tracy v. Dandurand, 30 S.W.3d 831 (Mo. 2000)

Lipton Realty, Inc. v. St. Louis Housing Authority, 705 S.W.2d 565 (Mo. App. E.D. 1986)

State ex rel. Shelter Mutual Insurance Co. v. Wagner, 575 S.W.3d 476 (Mo. App. W.D. 2018)

Diehl v. Fred Weber, Inc., 309 S.W.3d 309 (Mo. App. E.D. 2010)

State ex rel. Mitchell Humphrey & Co. v. Provaznik, 854 S.W.2d 810 (Mo. App. E.D. 1993)

Missouri Rules

Mo. Ann. Stat. § 491.060(3)

Missouri Supreme Court Rule 56.01

Missouri Rule of Professional Conduct 4-1.6