Photo of Michelle Yeary

Back in 2006, people were still carrying around Motorola Razrs, “YouTube” was barely a year old, and nobody had heard of an iPhone. That is when the device in Aceste v. Stryker Corp., 2026 Ohio Misc. LEXIS 1215 (Lucas Cty. C.P. Feb. 4, 2016),was implanted. By the time the case was filed in 2015, smartphones were ubiquitous and people were binge-watching Making a Murderer on Netflix. Now here we are in 2026, and after all that time, plaintiff has accomplished almost nothing. Summary judgment was a complete defense victory on every claim and on multiple independent grounds—causation, preemption, and statute of limitations.

The case involved an investigational spinal implant device subject to an IDE (investigational device exemption). Plaintiff alleged two injuries: (1) excessive wear debris allegedly causing metallosis or metal poisoning, and (2) fusion of the device that prevented it from articulating as designed and caused pain. Id. at *6-7.

Oddly, an Ohio trial court was applying Florida law to the substantive claims, but its causation analysis was spot on. And plaintiff’s proof problems were glaring. Plaintiff had no evidence whatsoever that he suffered metallosis or metal poisoning. Id. at *6. Medical records did show that the device had fused. Id. at *7. But identifying an injury is only half the causation equation. Plaintiff still had to prove that a defect in the device caused that injury.

That is where the case fell apart. Plaintiff offered testimony from a metallurgical engineer opining that the device suffered from manufacturing defects that could increase the production of metal debris inside the body. But the expert said nothing about whether those alleged defects could cause the device to fuse or fail to articulate. Id. at *7-8. In other words, plaintiff had evidence of a defect that could cause an injury he did not have, but no evidence of a defect that could cause the injury he did have. That mismatch proved fatal. Without evidence either that plaintiff suffered from excessive metal debris or that the fusion/non-articulation was caused by a manufacturing defect, there was no triable causation issue.

Plaintiff tried to bridge the evidentiary gap with a res ipsa loquitur argument. The device was explanted, therefore it must have been defective. The court rejected that theory outright as plaintiff’s statements were simply argument. Id. at *8. “[T]he fact that a course of treatment was unsuccessful does not alone prove a manufacturing defect.” Id. at *9. Even assuming counsel was correct that “many” of these devices allegedly failed to meet specifications (counsel claimed to have other clients with the same device), plaintiff still lacked evidence that his particular device deviated from specifications or that any such deviation caused his injuries. Res ipsa could not substitute for proof.

Even though the causation ruling alone disposed of the case, the court went on to address preemption—and delivered another sweep for defendants.

Plaintiff relied heavily on Mink v. Smith & Nephew, Inc., where the Eleventh Circuit allowed the plaintiff’s manufacturing defect claims to proceed as “parallel claims” because Florida recognizes a strict product liability claim based on a manufacturing defect and the plaintiff alleged that the defendant “violated the Florida common law duty to use due care in manufacturing a medical device.” Now, we think the Eleventh Circuit got it wrong because the supposed “parallel” duty in that case boiled down to allegations that the manufacturer violated FDA requirements—a classic attempt at private FDCA enforcement dressed up as state tort law. But that’s neither here nor there for Aceste because Mink was about a PMA medical device and Aceste involves an IDE device. That distinction mattered. IDE devices are still in the clinical trial phase gathering data for the PMA process. As the court recognized, IDE devices are not subject to the same sort of detailed, device-specific federal requirements that plaintiffs often try to use as the basis for parallel claims. And even if Mink applied, plaintiff here never identified any specific federal requirement that defendants allegedly violated. Without an identified federal requirement to parallel, there was no escaping preemption. The court therefore held all claims preempted. Id. at *15-16.

The component supplier defendants had an additional layer of protection. Claims against them were barred by the Biomaterials Access Assurance Act (“BAAA”), which generally shields component suppliers where the manufacturer accepted the component parts. Yet another dead end for plaintiff. Id. at *16-19.

Finally, the court addressed statute of limitations. There was evidence plaintiff knew as early as January 2007 that the device had fused and was causing him pain. Plaintiff nevertheless argued that the statute should not begin running until 2015, when he allegedly first learned of a potential manufacturing defect. But Ohio’s discovery rule applies to latent injuries, not latent defects. The relevant inquiry was when plaintiff had constructive knowledge of the injury and its factual cause—not when he later learned of a possible legal theory. Id. at *20-23. Once plaintiff knew the device had fused and was allegedly causing pain, the clock started ticking. By the time suit was filed, Ohio’s two-year statute of limitations had long expired.

After more than ten years of litigation over a device implanted nearly twenty years ago, plaintiff still could not produce evidence tying any alleged manufacturing defect to his actual injuries, could not avoid preemption, and could not overcome the statute of limitations. The result was a clean sweep for defendants across the board.

Photo of Lisa Baird

In the drug and device product liability world, we love our acronyms and our short-hand phrases.  The MDAs to the FDCA.  Class III.  PMA.  510(k).

Today’s acronym is CGMP, which sometimes you will see written as “cGMP”.  The GMP stands for Good Manufacturing Practices, and the “C” (or “c”) has, since a 1996 Final Rule—and until recently—stood for Current. 

These CGMPs have been housed in the Code of Federal Regulations, 21 C.F.R. 820, et seq. (which, to add to the acronyms, are sometimes called “QSRs” for Quality System Regulations or “QMSR” for Quality Management System Regulations).

Although the CGMPs were meant to “govern the methods used in, and the facilities and controls used for, the design, manufacture, packaging, labeling, storage, installation, and servicing of all finished devices intended for human use” and “intended to ensure that finished devices will be safe and effective and otherwise in compliance with the Federal Food, Drug, and Cosmetic Act (the act)” [21 C.F.R. § 820.1(a) (1996)], they were never prescriptive.

They were meant to be tailored by manufacturers to the particular device at issue, and they were meant to evolve and adapt to fit the circumstances.  

For example, 21 C.F.R. § 820.22 (available at https://www.govinfo.gov/content/pkg/FR-1996-10-07/pdf/96-25720.pdf) required device quality checks, but it provided:

Each manufacturer shall establish procedures for quality audits and conduct such audits to assure that the quality system is in compliance with the established quality system requirements and to determine the effectiveness of the quality system.

Similarly, 21 C.F.R. § 820.70 required manufacturing processes that would prevent non-conforming products, but stated:

Each manufacturer shall develop, conduct, control, and monitor production processes to ensure that a device conforms to its specifications.

Design, manufacture, labeling, adverse event reporting—all the regulations followed the same pattern.  Manufacturers need to have processes covering these areas, but those processes are not “one size fits all.”

Over the years, many plaintiffs have tried to plead around express preemption for Class III, PMA devices [21 U.S.C. § 360k(a) as interpreted in Riegel v. Medtronic, Inc., 552 U.S. 312 (2008)] by citing a laundry list of allegedly-violated CGMPs.  

The courts that have properly understood the non-prescriptive aspect of alleged CGMP violations have found such allegations insufficient to support a claim that escapes preemption because these CGMPs always have been “intentionally vague and open-ended.” Ilarraza v. Medtronic, Inc., 677 F. Supp. 2d 582, 588 (E.D.N.Y. 2009); see also In re Medtronic, Inc. Sprint Fidelis Leads Prod. Liab. Litig., 592 F. Supp. 2d 1147, 1157 (D. Minn. 2009) (CGMPs “are inherently flexible” and “require manufacturers to develop their own quality-system controls”).

A more compelling reason exists for preemption of CGMP-based claims, however.  For devices approved through the PMA process, the FDA actually reviews and approves the manufacturer’s quality management system processes as part of the PMA process and the FDA then is charged with enforcing what has been transformed into device-specific federal requirements.

As the FDA explained in “Quality System Information for Certain Premarket Application Reviews; Guidance for Industry and FDA Staff” (available at https://www.fda.gov/media/71083/download),

A Premarket Approval Application (PMA) is required to include a complete description of the methods, facilities, and controls, in sufficient detail so that FDA can make a knowledgeable assessment of the quality control used in producing the medical device (21 U.S.C. 515(c)(1)(C)).

Not enough?  See also FDA’s Compliance Program Guidance Manual related to Medical Device PMA Preapproval and PMA Postmarket Inspections, at Part I, p. 1-2:

PMA applications “are required to include” descriptions of their compliance with Current Good Manufacturing Practices (“CGMP”) requirements, which are promulgated in the Quality System Regulation (“QSR”), and “approval of a PMA application for a device can be denied if a manufacturer does not conform to the QS regulation requirements.”  

In other words, although all devices have to have tailor-made quality processes as outlined in the otherwise vague regulations at  21 C.F.R. § 820 et seq., for PMA devices, the FDA always has had the last word on whether those processes were appropriate and adequate—and once the FDA mandated those processes by granting premarket approval, changes to those devices that might affect safety or effectiveness had to be approved by the FDA as well.  

Accordingly, although we certainly agree that superficial citations to various CGMP regulations have always failed to assert a non-preempted claim because the regulations themselves are too vague and open-ended, that is not even half of the preemption story.  

The rest of the preemption story is that, as part of the PMA process, the FDA actually examines the manufacturer’s proposed processes and then mandates (through premarket approval) what those processes must entail.  A tort plaintiff who tries to use state law to fault the FDA-required process and suggest that a different one should have been used is, by definition is seeking to enforce a state law requirement that is different from, or in addition to, what the FDA itself required as a matter of federal law.   

The reality of the FDA’s regulatory processes matter to preemption, and it is important to inform courts about those processes because they are not well-developed in the preemption case law.  

Please also note that the CGMPs of 21 C.F.R. 820 et seq. themselves recently underwent a significant change, effective February 2, 2026.  In 89 Fed. Reg. 7496, the FDA issued

a final rule to amend the device current good manufacturing practice (CGMP) requirements of the Quality System (QS) regulation to harmonize and modernize the regulation. We are harmonizing to align more closely with the international consensus standard for devices by converging with the quality management system (QMS) requirements used by other regulatory authorities from other jurisdictions (i.e., other countries). We are doing so by incorporating by reference an international standard specific for device quality management systems.

The main international standard:  ISO 13485 (2016), Medical Devices—Quality Management System Requirements for Regulatory Purposes.  (Users who register can download a read-only version of ISO 13485 at https://ibr.ansi.org/Standards/iso2.aspx ).

This change-over—from federal regulations covering the various quality system topics to incorporation of ISO 13485 which does much the same thing—should not alter the preemption arguments.  

Like the CGMPs before it, ISO 13485 requires the development of device-appropriate design, manufacture, labeling, and other processes, and relevant for preemption purposes, those will still require FDA review and approval through the PMA process, and that should mean preemption is the result.  

Photo of Eric Hudson

About a year ago we blogged about a strong preemption decision from the Eastern District of New York, Gallego v. Tandem Diabetes Care, Inc., 2025 WL 948282 (E.D.N.Y. March 28, 2025). Gallego involved a Class III insulin injection pump that sent a warning alarm to its user that insulin was no longer being delivered. The diabetic using the pump then spoke with a representative of the manufacturer who confirmed that the pump was not delivering insulin and that the user should replace the insulin cartridge. The user of the pump died later the same day, and his estate filed a lawsuit. As discussed in our prior post, the court dismissed with prejudice most of the estate’s claims, but it granted plaintiff leave to file a second amended complaint (SAC) that included a claim for negligent defective design and wrongful death. Today’s decision, Gallego v. Tandem Diabetes Care, Inc., 2026 WL 1130316 (E.D.N.Y. Apr. 27, 2026), applies another robust preemption analysis and dismisses the plaintiff’s remaining claims with prejudice.

Continue Reading Preemption Round Two in the Eastern District of New York
Photo of Bexis

Bexis has been working with Lawyers for Civil Justice on a number of projects, including the currently pending initiative to enact a federal rules amendment that requires meaningful disclosure of third-party litigation funding (“TPLF”) on essentially the same rationale that insurance policies are routinely disclosed under Fed. R. Civ. P. 26 (a)(1)(A)(iv).  As part of this project, LCJ has its own initiative – “Ask About TPLF” – to encourage defendants, including defendants in prescription medical product liability litigation, to seek discovery of TPLF.

If you or your clients are participating in this LCJ initiative, or otherwise involved in seeking TPLF discovery, then you know that the other side almost always seeks to impede TPLF discovery with questionable claims of attorney/client privilege or work product protection.  Well, here’s something new that can help overcome such objections.

On April 7, 2026, Todd Presnell, a partner at Bradley Arant, submitted a comment to the Committee on Rules of Practice and Procedure (the judicial committee considering the TPLF rules change), which “conclude[d] that mandating disclosures will not adversely affect a party’s attorney-client privilege shield for communications or that party’s or her lawyer’s work-product.”  Presnell comment at 1.

Why should we care?

Because Presnell has probably forgotten more about attorney/client privilege/work product protection than any of us here on the Blog know.  As stated in the comment:

I have devoted a substantial amount of my practice to the study and analysis of . . . the attorney-client privilege and work-product doctrine. . . .  I have published over 40 articles in various journals on privilege-related issues. . . .  Separately, I am the sole creator and author of the legal blog Presnell on Privileges . . . and the lead author of the legal treatise Privileges and Protections: Tennessee and Sixth Circuit Law. . . .  I have given over 100 legal-education presentations or client-focused privilege training sessions, been retained to author and file several amicus briefs on privilege issues, . . . and been retained by law firms to serve as a legal expert witness in the area of evidentiary privileges.

Id. at 1-2.  This guy knows of what he speaks.

And what did he have to say?

In general, and with plenty of citations to back it up, that neither the attorney/client privilege nor the work product protection is an impediment to a rule requiring the general disclosure of TPLF contracts.  Id. at 2.  Other documents, maybe, but not the TPLF contracts themselves.  Why?

A TPLF Agreement between a party, or that party’s counsel, and a non-party funding entity is not a communication, not a communication between a client and her lawyer, and does not pertain to the request for or provision of legal advice. . . .  A TPLF Agreement arises in the ordinary course of a non-party funding entity’s business.

Id. 2.  In short, privilege-based objections to the proposed TPLF disclosure rule are overblown and should be limited to “putatively protected portions” of the actual contracts.  Id.

Here is an outline of the rest of the privilege argument:

  • Because privileges preclude discovery of otherwise relevant information, they are interpreted narrowly.  Id. at 3.
  • The attorney/client privilege is limited to confidential communications between clients and attorneys for the purpose of obtaining legal advice.  Id. at 3-4.
  • Even where applicable, the privilege can be waived.  Id. at 4.
  • A TPLF agreement is a contractual agreement, not a communication.  Id.
  • A TPLF contract cannot be confidential because it necessarily involves a third-party funder.  Id.
  • A TPLF contract does not involve provision of legal advice.  Id.
  • If a TPLF contract includes an otherwise privileged communication, that is a waiver of confidentiality.  Id.

The work product doctrine is different.  It protects information an attorney uses in anticipation of litigation that either the attorney or the client prepares.  It does not extend to material prepared in the attorney’s ordinary course of business.  Id. at 5-6.  The work product protection does not extend to a TPLF contract because that kind of agreement “arises [from] the funder’s decision to invest capital as part of its ordinary business and [the] attorney’s need for business financing.”  Id. at *7.  This section is notable for its comprehensive discussion of relevant precedent, as to both TPLF contracts and “other litigation-related agreements” such as contingent fee documents, invoices, and billing statements.  Id. at *7-8.  It concludes that “[a] TPLF Agreement does not warrant work-product protection because, while related to litigation, it is a [contract] created in the ordinary course of the non-party funder’s business for the purpose of memorializing a financing arrangement.”  Id. at *8.

So if you’re involved in motion practice concerning production of TPLF agreements, take a look at Presnell’s public comment.  Unless and until the rules committee acts, it can save time and money while increasing your understanding of why the other side’s privilege arguments can’t hold water.

Photo of Michelle Yeary

If Rule 702 is supposed to keep unreliable expert opinions out, this decision raises a fair question: how many gaps are too many? In In re Covidien Hernia Mesh Products Liability Litigation, 2026 WL 1129617 (D. Mass. Apr. 27, 2026), the court considered a familiar lineup of expert opinions—general causation, specific causation, alternative design, and warnings. The outcome is mixed. Most of plaintiff’s expert’s opinions survive. But not all.

Plaintiff underwent emergency hernia repair surgery and received defendant’s surgical mesh, a polyester textile with a porcine collagen and glycerol coating designed to prevent tissue attachment during healing. Three years later, plaintiff required another surgery to repair a recurrent hernia and bowel obstruction. During that procedure, the surgeon found a portion of plaintiff’s bowel “stuck tightly” to the prior mesh. Id. at *3.  Plaintiff’s theory, offered via a general surgery expert, is that the polyester triggers a chronic inflammatory response, and the collagen barrier—represented to last “less than 1 month”—actually degrades too rapidly, allowing adhesions to form and leading to complications like obstruction and recurrence. Id. at *2-3.

The primary attack on general causation focused on the expert’s reliance on animal studies while largely excluding human clinical data. And while the court acknowledged that opinions that rely on animal studies should be “scrutinized” due to the difficulty in extrapolating to humans, it allowed it here, especially given the ethical limitations on conducting human studies:

testing the barrier’s resorption in humans in the period immediately following a hernia repair would require reopening patients. Given this ethical limitation, animal studies and anecdotal case reports are a reliable basis for reaching a scientific conclusion about the resorption period of the barrier.

Id.  at *8.

That’s a fairly narrow, case-specific ruling which was bolstered in the court’s opinion by the fact that the studies the expert relied on were conducted by the defendant itself. Even though that says little about whether the expert reliably extrapolated from those studies to his ultimate conclusions.

The more notable move comes next. The court admitted that plaintiff’s expert lacked human clinical data on inflammatory response and could not clearly explain how his animal-based conclusions translated to humans. Id. at *9. Ordinarily, that would be a problem for the proponent of the testimony. Here, it wasn’t. Because defendant did not cite product-specific clinical studies affirmatively disproving the expert’s theory, the court allowed the opinion to stand. Id. That framing effectively flips the burden. Instead of requiring plaintiff to demonstrate reliability, the court faulted defendant for not filling the gaps.

The same dynamic appears in the court’s treatment of contrary clinical data. Defendant argued the expert cherry-picked his data by ignoring product-specific studies that showed no increased rate of relevant complications. Because plaintiff’s expert dismissed those studies as having “limitations,” the court found defendant’s argument went to weight, not admissibility. Leaving cross-examination—rather than Rule 702—to do the work. Id.

The court also admitted the expert’s specific causation opinion based on a differential diagnosis. On the “rule in” side, the expert pointed to the presence of “dense” adhesions observed during the explant surgery years later as support for his theory that adhesions began forming shortly after implantation due to premature barrier degradation. That was enough for the court. Id. at *10.

On the “rule out” side, the burden of proof was again flipped. Defendant identified two obvious alternative causes: the urgent nature of the surgery and the surgical technique used—both known risk factors for recurrence. But instead of requiring the expert to meaningfully rule them out, the defendant was required to “offer data” to rule them in.  Id. at *11. The result is a differential diagnosis that does not do much ruling out, yet still survives.

If the court was permissive on causation, it was not on design defect. Plaintiff’s expert offered two purported safer alternatives: a different material and a different product. That was essentially it. No comparative analysis. No discussion of risks and benefits. No methodology tying the alternatives to a safer outcome. That, the court held, is classic ipse dixit—and inadmissible. Plaintiff’s attempt to rehabilitate the opinion through attorney argument in briefing came too late and in the wrong form. Without support in an expert report, deposition, or affidavit, the expanded theories could not be considered. Id. at *12-13.

Finally, the court allowed the expert to testify about warning adequacy from a surgeon’s perspective. A practicing physician can opine on whether warnings communicate sufficient information for clinical decision-making. But the expert cannot offer legal conclusions about adequacy under the law. Nor is he required to opine on warning causation—that is, whether a different warning would have changed the treating surgeon’s conduct. Id. at *13-14.

Taken as a whole, the decision reads less like strict gatekeeping and more like gap tolerance. The court repeatedly acknowledges analytical shortcomings—missing data, unexplained extrapolations, incomplete differential diagnosis—but treats them as fodder for cross-examination rather than grounds for exclusion. In doing so, it edges toward placing the burden on defendants to disprove opinions that plaintiffs have not fully substantiated. Fortunately, the alternative design ruling cuts the other way. However far plaintiff’s causation opinions are allowed to go, design defect claims still require evidence of a feasible, safer alternative supported by a reliable methodology. Without that cornerstone, plaintiff’s design defect theory may have a short shelf life.

Photo of Bexis

We’ve never done a cross-post before with any other Reed Smith (or any other firm’s) blogs, but the recent decision, Louisiana v. FDA, ___ F.4th ___, 2026 WL 1194924 (5th Cir. May 1, 2026), justifies this unprecedented action. Essentially, the Fifth Circuit, ordered a nationwide injunction against the FDA’s 2003 REMS amendments that authorized prescription of mifepristone by telemedicine – thereby requiring in person prescriptions that the FDA had determined were unnecessary, given the drug’s safety. With the change of administrations, the FDA is now controlled by RFK Jr. and has switched sides, with little scientific explanation, and no longer defends the REMS. As we know, the Supreme Court frowned on nationwide injunctions in Trump v. CASA, 606 U.S. 831 (2025). That proved of little consequence to the Fifth Circuit, which disposed of that argument in a single sentence, that the Supreme Court didn’t prohibit such injunctions under the Administrative Procedure Act. La. v. FDA, 2026 WL 1194924, at *9. So that means, in the Fifth Circuit, the APA now enjoys a more exalted position than the US Constitution itself, since CASA involved constitutional issues. That’s one of many topsy-turvy things about this ideologically driven decision.

Without further ado, here is the cross post from Reed Smith’s Health Industry/Washington Watch blog, authored by Scot Hasselman, Lesley Reynolds, Sarah Cummings Stewart, Allie Klimkiewicz, Alexa Chronister, and Matt Loughran reprinted in full.

**********

On Friday, May 1, 2026, a unanimous three-judge panel of the U.S. Court of Appeals for the Fifth Circuit granted the State of Louisiana’s emergency motion for a stay pending appeal in State of Louisiana v. FDA, No. 26-30203. The order immediately reinstated the in-person dispensing requirement for mifepristone, which was previously rolled back in 2021 (and formalized in 2023) by the FDA’s Risk Evaluation and Mitigation Strategy (REMS). In plain English: health care providers can now only prescribe mifepristone in-person and may no longer prescribe it virtually via telehealth. However, early on Monday, May 4, 2026, the Supreme Court issued an administrative stay of the Fifth Circuit’s order until May 11, effectively restoring the status quo for at least a week. This blog post analyzes the Fifth Circuit’s order and actions companies should consider during this interim time.

Mifepristone is used in most medication abortions in the United States (in a regimen with misoprostol); it is also widely used in miscarriage care. This ruling does not impact the FDA’s approval of mifepristone, though it may create legal and practical obstacles to accessing mifepristone. For example, patients may face barriers and longer wait times to pick up the medication in person, even in states where abortion remains legal, and even when mifepristone is being used for non-abortion purposes, like miscarriage management. This case is one of several actions brought by states seeking to prohibit telehealth and remote dispensing of mifepristone to patients in states with abortion bans, and the outcome here may influence those parallel proceedings. 

Friday’s ruling overturned the Western District of Louisiana’s April 2026 decision, which refused to stay the 2023 REMS requirements. There, the district court held that while the State of Louisiana demonstrated standing, a strong likelihood of success, and irreparable harm, the balance of equities and public interest weighed against Louisiana. Instead, the Western District of Louisiana stayed the entire case to allow FDA to complete its ongoing comprehensive review of the mifepristone REMS. On appeal, the Fifth Circuit held that the district court abused its discretion when balancing the equities and public-interest, as the Fifth Circuit emphasized that the likelihood of success and irreparable harm factors are the most critical factors upon which to focus.

In granting the administrative stay, Justice Alito did not evaluate the merits of either the district court’s or the Fifth Circuit’s holdings. Rather, the 7-line Order merely granted the stay and set deadlines for a response and expiration of the stay.

Background

Mifepristone was approved by FDA in 2000 for use in combination with misoprostol to end early pregnancy and for early miscarriage management, and has been subject to successive REMS modifications. A REMS is a drug safety program that the FDA can require for certain medications with serious safety concerns to help ensure the benefits of the medication outweigh its risks.

When initially approved, the REMS imposed significant access restrictions, which included three in-person visits and mandatory reporting of serious adverse events. Over time, the barriers to access mifepristone were progressively lowered. In 2016, the FDA allowed nurse practitioners to prescribe mifepristone after one in-person visit and limited reporting to only fatalities. During the COVID-19 pandemic, the FDA issued a Non-Enforcement Decision, by which FDA removed the in-person dispensing requirement entirely, permitting mail-order and pharmacy dispensing without an in-person visit, which were later formalized in the 2023 REMS.

In October 2025, Louisiana challenged the 2023 REMS under the Administrative Procedure Act (APA), arguing that FDA’s justifications for permitting remote dispensing relied on flawed or nonexistent data and that the 2023 REMS changes led to illegal abortions in Louisiana and unrecoverable Medicaid costs for alleged complications. Danco Laboratories and GenBioPro (the brand and generic manufacturers of mifepristone) intervened as defendants. The district court denied Louisiana’s motion for preliminary relief and stayed the litigation pending FDA review. Louisiana then appealed to the Fifth Circuit. 

The Fifth Circuit’s opinion begins by invoking the post-Dobbs landscape of state-level regulation and noting that President Biden’s Executive Order No. 14076 directed federal agencies to expand access to medication abortion. The Court also highlights FDA’s own recent concession—made when it announced its comprehensive review of mifepristone late last year—that the prior REMS approvals suffered from a “lack of adequate consideration.” Together, this contextual framing sets the stage for the Court’s decision to grant Louisiana’s emergency motion for a stay pending appeal, discussed further below. 

Key Holdings on the Stay Factors Under 5 U.S.C. § 705

Louisiana sought a stay of the 2023 REMS under 5 U.S.C. § 705, which authorizes courts to “postpone the effective date of an agency action or to preserve status or rights pending conclusion of the review proceedings.” The Court applied the standard four-factor test as follows:

  1. Strong Likelihood of Success. The Court held Louisiana is likely to prevail on its APA claim, calling the scientific literature supporting the 2023 REMS modification inadequate and a “textbook example of arbitrary and capricious agency action.” 
  2. Irreparable Harm. The Court held Louisiana has shown that it is irreparably harmed without a stay, pointing to Louisiana’s argument that the 2023 REMS undermines Louisiana’s abortion laws and causes financial harm to Louisiana from Medicaid expenditures.
  3. Equities and Public Interest. The Court agreed with Louisiana that the district court erred by finding Louisiana’s irreparable harms are outweighed by FDA’s interest in continuing its review and Danco’s financial interests in selling mifepristone. The Court held that “[n]either the FDA nor the public has any interest in enforcing a regulation that violates federal law,” noting this was the third time the Fifth Circuit has found that FDA’s relaxation of mifepristone’s safeguards “likely lacked a basis in data and scientific literature.” The Court emphasized that FDA itself concedes the 2023 REMS was marred by “procedural deficits” and a “lack of adequate consideration,” and that “the public interest is not served by perpetuating a medical practice whose safety the agency admits was inadequately studied.”

The Fifth Circuit’s order is applied nationwide, goes into effect immediately (subject to the Supreme Court’s temporary administrative stay), and does not disturb the underlying FDA approval of mifepristone itself or the continued review that the FDA is conducting. The Court cautioned that while this ruling has nationwide effect, it should not be construed through the same lens as other nationwide injunctions given that this is a remedy of the Administrative Procedure Act, and therefore appropriate. In other words, though this ruling functions practically as a nationwide injunction, it does not violate the Supreme Court’s 2025 limitations on universal injunctions, as the § 705 stay “temporarily voids the challenged authority,” and here, the challenged authority is a nationwide regulator action.

Key Implications

For pharmaceutical manufacturers, pharmacies, telehealth platforms, health systems, distributors, and other stakeholders, Friday’s order creates immediate compliance and operational obligations under the reinstated, pre-2023, REMS. Though the Supreme Court has temporarily issued an administrative stay, companies should nonetheless consider the below recommendations in the event the stay is soon lifted:

  • REMS Compliance and Enforcement Risk: Unless the Supreme Court’s administrative stay is extended, certified pharmacies and distributors will need to immediately revert to in-person-only dispensing protocols for mifepristone. Mail-order and telehealth-based fulfillment of mifepristone prescriptions would be prohibited. Manufacturers should be prepared to promptly update REMS materials, labeling, distribution agreements, and training to reflect the pre-2023 framework. 
  • Telehealth and Virtual Care Disruption: Unless the Supreme Court’s administrative stay is extended, providers and platforms offering medication abortion or miscarriage management involving mifepristone via telehealth must require in-person evaluations or clinic-based dispensing. This will disproportionately impact access in rural or restrictive jurisdictions and may create tension with state shield laws in permissive states. 

What’s Next?

On Friday night, mere hours after the Fifth Circuit issued its decision, Danco Laboratories, filed for an emergency one-week administrative stay of the Fifth Circuit’s order to allow Danco time to seek relief in the United States Supreme Court. The motion requested a decision by the Fifth Circuit by 9 pm on Friday. Shortly after that deadline passed, Danco then filed an application for an emergency stay to the Supreme Court. Applications from the Fifth Circuit go to Justice Samuel Alito for initial review, who can then either rule on the application individually or refer it to the full court for review. As of Monday morning, Justice Alito administratively stayed the Fifth Circuit’s order until 5:00 pm EDT on Monday, May 11, 2026. A response to the manufacturers’ application for a stay is due by 5:00 pm EDT on Thursday, May 7, 2026. 

Meanwhile, FDA’s comprehensive review of mifepristone, which began in September 2025 and will include consideration of the 2023 REMS, continues in the background, though the agency has not committed to a timeline for completing its review. The results of that review could also influence the ongoing proceedings in this and other cases.

While the Supreme Court’s temporary administrative stay has restored the status quo, within weeks it could lift that say, reinstating the in-person dispensing requirement in full effect. Notwithstanding the administrative stay, manufacturers, pharmacies, telehealth providers, and health systems should evaluate immediate operational changes and consider strategic engagement with FDA where appropriate. 

Reed Smith will continue to track developments in the post-Dobbs abortion landscape and the regulation of mifepristone in particular. If you have questions about how this decision affects your REMS compliance program, distribution practices, or telehealth operations, please reach out to the health care lawyers at Reed Smith. 

Photo of Steven Boranian

After nearly three years of litigation winding through California’s appellate courts, the closely watched Gilead Tenofovir Cases is finally set for oral argument before the California Supreme Court on Wednesday, May 6, 2026, at 9:00 a.m. in San Francisco. The case presents a question with potentially sweeping consequences for product liability law: Does a pharmaceutical manufacturer owe a duty of reasonable care to users of a non-defective medicine when making decisions about the development of an allegedly safer and equally effective alternative compound? Based on long-standing and well-grounded principles of tort law, the correct answer is that no such duty exists or should exist.

Of course, our view on this controversy is well known, having covered the case here, here, here, and here.  For the upcoming oral argument in the California Supreme Court, here is what you need to know. 

The dispute centers on Gilead Sciences and its development of live-saving and life-extending HIV antiretroviral medicines. The plaintiffs are patients who used medicines containing tenofovir disoproxil fumarate (“TDF”) and allegedly experienced warned-of side effects. A positive HIV test was once considered a death sentence, but TDF-based medicines, which the FDA first approved in 2001, changed that and have been credited with saving countless lives.

Most importantly, the plaintiffs do not allege that the TDF medicines they used were defective in design, manufacture, or labeling. That’s right, no alleged product defect.

Instead, the plaintiffs claim that the manufacturer was too slow in developing a different compound, tenofovir alafenamide (“TAF”), which they allege has a better safety profile. The FDA approved TAF-based products in 2015, but in the plaintiffs’ telling, the purported delay in developing these alternative therapies was unreasonable. Of course, in reality there was no “delay” at all, but merely a manufacturer’s decision to dedicate limited development resources to a product that ended up saving thousands of lives. Product manufacturers make product development decisions like these almost every day.

The trial court denied summary judgment on negligence, and the California Court of Appeal affirmed in an opinion that came in as our worst drug or medical device case of 2024. In a first-of-its-kind opinion, the Court of Appeal ruled that a pharmaceutical manufacturer has a duty of reasonable care when it has invented “what it knows is a safer, and at least equally effective, alternative to a prescription drug that it is currently selling and that is not shown to be defective.” Gilead Tenofovir Cases, 98 Cal.App.5th 911, 922 (2024).

The Court of Appeal basically purported to create a new tort—a claim sort of like negligence, but requiring “knowledge,” and custom designed to allow judges and juries to second guess isolated product development decisions, sometimes decades after the fact. This is all in the absence of any allegation of any product defect, let alone proof. Worse yet, the court drew an analogy to Mexicali Rose v. Superior Court, where the California Supreme Court held that a restaurant could be liable in negligence for a chicken bone found in an enchilada. Development of prescription medicines is not like cooking chicken enchiladas.

So here we are. The California Supreme Court granted the manufacturer’s petition for review, and oral argument is set for Wednesday, May 6, 2026, at 9:00 Pacific. The argument will be heard in the Supreme Court’s courtroom in San Francisco, with live-streaming available on the Supreme Court’s website. (Viewing California Supreme Court arguments can be a bit wonky, but if you scroll down to the Oral Arguments section and click through, you should be okay.)

Several critical issues are likely to dominate the argument. First, the Court will confront the threshold question of whether a negligence claim against a product manufacturer can proceed in the absence of any allegation that the product was defective. This question implicates the relationship between strict product liability and negligence, which have traditionally shared a common requirement of proving a product defect. That requirement goes back to the earliest days of product liability law in California and elsewhere.

Second, the Court will likely probe the boundaries of any duty it might recognize. The Court of Appeal attempted to characterize its holding as “narrow,” but we don’t buy it. The Court of Appeal’s attempt to define the purported duty was tortured at best, and the potential for confusion and slippery expansion clearly exists.

Third, public policy implications are obvious. In Brown v. Superior Court, 44 Cal. 3d 1049 (1988), the California Supreme Court rejected strict product liability for prescription drugs in part because public policy favors the development and marketing of beneficial new drugs. The Court of Appeal’s new duty directly undermines this precedent. Moreover, the Court of Appeal’s new duty potentially impacts not only pharmaceutical manufacturers, but also manufacturers in other industries driven by innovation. The International Center for Law & Economics argued in its amicus brief that the Court of Appeal’s rule effectively “imposes unlimited liability” by eliminating the defect requirement that has long served as a cornerstone of product liability law.

If the California Supreme Court affirms the Court of Appeal, it will establish a new theory of manufacturer liability that has no precedent in any American jurisdiction. If the Court reverses, it will reaffirm the centrality of the defect requirement to product liability law and retain potential negligence liability within well-established limits.

All eyes will be on San Francisco on May 6.

Photo of Bexis

Bexis recently attended the Spring Conference of the Product Liability Advisory Council (“PLAC”).  PLAC meetings are usually good for new blogpost ideas, and this one was no exception.  Today’s idea comes from an unusual source, though – the final day’s ethics presentation.  That presentation was about artificial intelligence, mostly in the mass tort context.  One segment pointed out how the other side is using AI for, among other things, new plaintiff intake.  That prompted us to take a look at who’s advertising what to the other side.  We don’t particularly want to promote their business, but an on-line journal article we found stated:

AI tools can be used in mass tort management to streamline case intake by automating data collection from online forms, emails, and calls.  It can drastically reduce manual entry errors and time.  They analyze plaintiff information in real-time to assess eligibility and match individuals to appropriate tort categories.

We didn’t have to look far.  An ad masquerading as a blog promised that it “excels in lead screening” and could “pre-qualify callers by asking about drug usage, medical diagnoses, and side effects.” Another article promised “intake” that would “engage every caller, capture and analyze every detail, and prioritize the highest-value cases.”  We found advertisements pushing prospective plaintiff “identification and enrichment” that promised to create MDL-specific matrices and to populate them with “scored” data that included familiar information:  product use, qualifying injury, and statute of limitations.  Another promoted “AI-based automated intake forms” that could “restore client information and verify it with official databases at the same time as it detects missing documentation.”  Still another was offering AI “fillable PDFs” with “tools to help structure the data such as checkboxes and drop-down lists.”  These would collect information about would-be plaintiffs’ “general demographics,” “medical background,” “proof of use or exposure,” and “proof of the injury being alleged.”  We didn’t have to dig past the second page of Google results.

Maybe these plaintiff AI tools are as good as advertised, maybe not.  But the fundamental problem remains – mass tort incentives have never (at least prior to settlement) favored careful early vetting of plaintiffs.  Instead, the other side’s modus operandi has always been “the more, the merrier.”  They get names and file claims, and then they leave it to the defense to spend the time and effort to separate the good claims (if any) from those that would never have been filed in one-off litigation.  We have serious doubts that even the most efficient AI changes those incentives much.

However, plaintiffs’ use of AI plaintiff intake tools provides the defense an opportunity to put new Fed. R. Civ. P. 16.1 to work.  Specifically, Rule 16.1(b) requires (unless an MDL court prohibits it) that the parties address “how and when” they “will exchange information about the factual bases for their claims” (emphasis original).  Note use of “will” rather than “whether” in this section.  That means it’s mandatory (unlike most of the rest of Rule 16.1).  We should seek to obtain the factual (not privileged) plaintiff “information” that their AI has collected.  If these systems work as well as advertised, that should be easy for the other side to produce, and maybe the cost of vetting the bodies they dig up can be more equitably apportioned in MDLs.

But that’s not all.  Equally important, is what “information” we might still have to collect because plaintiffs failed to utilize all the advertised bells and whistles that these AI systems claim to possess.  We’re not talking about particular AI prompts − especially anything that may be subject to work product privilege.  What we are suggesting is determining whether the other side is affirmatively disabling their own AI systems so they can continue to file factually unsupported lawsuits.  That would seem to be a rather blatant violation of Rule 11, and we hope the other side does not do this.  However, their track record on vetting their own supposed clients is not very good.

Our side’s Rule 16.1(b) statements thus should make an effort to ensure that the other side is actually using the early vetting capacity of their own AI programs.  Given the “file and forget” incentives that the plaintiffs’ side has in mass torts, it would not surprise us at all that some plaintiff firms do not use much of the informational functionality of their own AI.  Knowing that would be useful, both in determining potential apportionment of discovery expenses for information plaintiffs could have collected, but did not, and more generally to call the other side’s credibility into question early in the MDL process.

So that is our take home idea from PLAC.  Since the other side claims to be using sophisticated AI tools in mass tort plaintiff intake – we should ensure that they are actually using their available AI capacity to do what Rule 11(b)(3) supposedly required them to be doing all along:  pre-complaint verification that their client’s “factual contentions have evidentiary support.”

Photo of Eric Alexander

No, not that case.  The appeal of the FIFRA preemption rulings in glyphosate litigation is still pending.  This decision is Hencely v. Fluor Corp., No. 24-924, 2026 U.S. Lexis 1868 (U.S. Apr. 22, 2026), which involves the preemption of negligence claims against a military contractor in connection with a terrorist attack on the Bagram Airfield in Afghanistan.  In some ways, the Hencely case is worlds away from product liability claims related to medical products.  In other ways, it is close enough for this Blog to pay attention, especially if you go back to some of the factual details from the Fourth Circuit decision that the majority Supreme Court decision omitted.  We will presume to explain, although with some hardy initial disclaimers.  First, the underlying events in Hencely are tragic.  Our post should not be read to minimize those events or draw any connections to current U.S. military activities.  Second, although the Blog has addressed the government contractor defense in connection with product liability claims, we are not experts in government contracts or national defense law.  We leave for others the discussion of whether Hencely will have a chilling effect on the availability of contractors to fill roles in connection with Department of Defense or other national security functions, an issue the dissent raised.  Id. at 36.  Third, the alignment of the Court—Justice Thomas writing for a majority that included Justices Sotomayor, Kagan, Jackson, Gorsuch, and Barrett and Justice Alito writing a dissent in which Chief Justice Roberts and Justice Kavanaugh joined—could invite a vigorous discussion about inconsistent positions being taken compared to some of the big prior decisions we discussed on preemption and other subjects, such as here, here, here, and here.  Suffice it to say that the five male members of the Court tend to be on the same side in terms of federalism, originalism, and stare decisis.  Not in Hencely.

From the title and preceding paragraph, the outcome of Hencely should be clear:  The Supreme Court reversed the district court’s grant of summary judgment based on preemption and Fourth Circuit’s affirmance.  Plaintiff sued a defense contractor and its subsidiaries over injuries he sustained at Bagram in 2016 while an active duty Army member in connection with a suicide bombing he tried to stop.  Tragically, this was one of several attacks on the base over time, including by infiltrators.  The terrorist who caused the injuries in Hencely got on base through a U.S. military policy called “Afghan First,” although he was employed by the defendant contractor and under its supervision.  The U.S. military also supervised and screened the terrorist leading up to the attack, according to the Fourth Circuit and the dissent but not mentioned in the majority opinion.  Plaintiff brought his suit in the District of South Carolina and asserted negligence claims under South Carolina law.  Although the U.S. was not sued—presumably because of sovereign immunity—even without details from the complaint, we can say that it would have been fairly obvious from the start that this case would have to involve evidence about policies, actions, and decisions of the U.S. military.  That realization, in turn, should have raised major constitutional concerns, because states lack authority to weigh in on how the U.S. military operates a base in Afghanistan.  Somehow, Hencely got past the pleading stage before the defendant won summary judgment.  554 F. Supp. 3d 770 (D.S.C. 2021).  The Fourth Circuit affirmed in a long and detailed opinion.  120 F. 4th 412 (4th Cir. 2024).  [For some reason, both the majority opinion and dissent in Hencely are littered with citation errors, which we will attempt not to repeat.]

We can cut to the chase and say that the Fourth Circuit and dissent made more sense and appeared to consider more of the record than the selective recounting in the majority decision.  Rather than go through the details of how the dueling opinions got to their respective conclusions, we want to focus on some of the language used and some of the omissions compared to the preemption analyses we usually see in drug and device cases. 

First, the Bexis bugaboo, the purported presumption against preemption.  There is no discussion of this concept in the majority decision or the Fourth Circuit’s decision.  The dissent has a discussion of how a presumption applies to field preemption in an area traditionally occupied by states but not to areas that the constitution or congress specified as federal.  Id. at *27.  Frankly, the majority could have referred to such a presumption as part of its rationale, but the omission may only matter to us.

Second, while we use the term “field preemption,” the majority and dissent in Hencely do not.  The majority uses the term “battlefield preemption” once when referring to the basis for the Fourth Circuit’s affirmance.  Id. at *9.  Five circuits, including the Fourth, have adopted that doctrine, and most press about Hencely has referred to that as the issue being decided, but the dissent never uses the term as it rebuts the majority’s rejection of it.  (The majority also does not give a tally of which circuits have or have not adopted the doctrine.)  The terms “express preemption” and “implied preemption” appear once each, both in the dissent in characterizing holdings in other cases.  No mentions of “conflict preemption” or “obstacle preemption” can be found.  This differs from our experience, which has typically involved analyzing which particular species of preemption applies to the asserted claims.  Of field, express, and implied, the last has a few subspecies.  Such labeling may be less important to this Court.  For instance, the distillation of the reasoning of the dissent’s position is decidedly unlabeled:

In sum, we have long recognized that the Constitution itself may demand preemption when a state law intrudes upon an area of exclusive federal authority. And because the Constitution gives the Federal Government exclusive authority over foreign affairs and the conduct of wars, federal law preempts all state law that substantially interferes with the Government’s exercise of those powers.

Id. at *28.

Third, the only discussions of a legal “defense” were in connection with the government contractor defense, which was not at issue on appeal, and a possible defense to punitive damages, which was not at issue on appeal.  The burden in connection with the motion below was not discussed.  The standards for summary judgment and the appellate review of a summary judgment decision are absent.  Whether the evidence in the record below left any genuine dispute as to material facts was not addressed, even to acknowledge any conflicting evidence.  The only mention of evidence was in the dissent in the context of pointing out that scrutiny of the U.S. military’s decisions and actions would be inevitable if the case went to trial.  This is also strange to us.  If anything, we might infer that the majority reached its decision based on the idea that it would be unfair to deprive plaintiff of a chance at recovery.  Similarly, we could offer that the dissent was driven by concerns about the imposition on the military of a case that second-guessed its policies and actions.  If we are right about the majority’s motivations, then that is not a great sign for medical product manufacturers up against sympathetic plaintiffs.

Fourth, in addition to the expected range of amici on each side, the United States submitted a brief in support of the defendant.  The Solicitor General was also permitted to participate in the oral argument.  It makes sense that the U.S. would favor preemption in a case where the military would otherwise be subject to burdensome and tricky discovery and be the subject of extensive second-guessing at trial.  Similar logic would suggest that the U.S. would support preemption in cases that fundamentally second guess the decisions of FDA, including its regulations.  Yet, U.S. amicus support in those cases has not been consistent.  In Hencely, the position of the U.S. did not sway the majority.  To the contrary, the report of the military finding fault with the contractor in connection with the terrorist attack featured much more prominently.  In contrast to the dissent, the majority largely ignored that the Afghan First program and the security requirements the military imposed would be major focuses in the case—which clearly supports preemption.  This reminds us of how some courts that allow preempted claims and evidence tend to focus on theoretical ways that the manufacturer could have acted differently while turning a blind eye to how plaintiffs hinge their case on the alleged inadequacies of the FDCA, FDA regulations, and FDA actions.  The Supreme Court should be more savvy about such things.

Photo of Bexis

Today’s guest post is another tech-related discussion from Reed Smith‘s Jamie Lanphear. Given the increasing ubiquity of artificial intelligence (“AI”) in legal practice, the notion of AI prompts and output becoming yet another front in the never-ending ediscovery wars is concerning. Here are Jamie’s latest thoughts on the latest pertinent caselaw in this area. As always our guest posters deserve 100% of the credit (and any blame) for their efforts.

**********

The Blog has recently covered both the discoverability of AI prompts (here) and the Heppner decision out of the Southern District of New York (here). Since those posts, courts have continued to develop the law in this area — and some of the developments are encouraging. This post picks up where those earlier discussions left off, surveying the latest case law on privilege, work product, and waiver as they apply to AI prompts and outputs, and offering practical guidance for lawyers and in-house teams navigating an area where the rules are still being written.

Heppner: A Quick Recap and What It Leaves Open

As the Blog discussed here, in United States v. Heppner, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026), Judge Rakoff held that a criminal defendant’s communications with a generative AI platform were protected by neither attorney-client privilege nor the work product doctrine. The holding rested on several narrow facts: the defendant used a public AI tool on his own, without direction from counsel; the platform’s terms of service destroyed any expectation of confidentiality; and the materials were seized pursuant to a search warrant in a criminal case rather than sought through civil discovery. The court did not apply or analyze work product protection under Rule 26(b)(3) of the Federal Rules of Civil Procedure. The decision generated significant concern – but the limited application left a number of important questions unanswered.

Open Questions: Privilege in the Corporate AI Context

Heppner is what we have on privilege, but it does not tell us very much about how privilege will apply in the corporate context. The case involved a party acting on his own, not for the purpose of seeking legal advice from counsel. The in-house world presents meaningfully different scenarios.

Start with the threshold problem. Attorney-client privilege requires a communication between an attorney and a client, and when a corporate employee communicates with an AI chatbot, the communication is between the employee and a machine — not between the employee and counsel.

However, there are more-than plausible arguments that some of these interactions should remain protected. If an attorney directs an employee to use an in-house AI tool as part of the process of providing legal advice, the tool looks more like an agent than a third party — functionally similar to a paralegal or other intermediary acting at counsel’s direction. No court has endorsed that theory yet, but it is not an unreasonable application of existing doctrine to novel facts. Thus, in the first place, attorneys should think long and hard before directing clients (rather than their own offices) to do something like this.

The same logic applies in more routine scenarios. If an employee communicates with a colleague to gather information at counsel’s request for the purpose of providing legal advice to the company, those communications are privileged. What if the employee consults an AI tool instead of a colleague? The substance of the interaction is the same — the employee is gathering information counsel needs to provide legal advice. The only difference is the medium.

A related question arises with enterprise AI tools. If a company deploys an internal AI system and counsel interacts with that system to obtain information necessary to provide legal advice, there is a reasonable argument that the tool functions as the client’s agent. That scenario is not meaningfully different from an attorney communicating with a client’s accountant or other third-party agent to obtain information needed to render legal advice. Enterprise tools are deployed by the company, access-controlled, and integrated into the legal workflow. They are not consumer-grade chatbots with permissive terms of service. That factual distinction was critical to the outcome in Heppner.

None of these theories has been tested. But it is unlikely that courts will fail to find ways to adapt privilege doctrine to accommodate the realities of AI use. The law may lag science, but it does not ignore technology. The analogies are there, and courts have adapted privilege to new contexts before. The questions are not whether privilege will evolve, but how, and with what nuances.

Morgan: A Counter to Heppner

A recent decision out of the District of Colorado offers a notably different perspective on whether AI interactions are protected — at least under the work product doctrine. In Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB, 2026 WL 864223 (D. Colo. Mar. 30, 2026), an employment discrimination case, a pro se plaintiff used AI in connection with litigation preparation. The defendant moved to compel disclosure of the AI tool the plaintiff was using and sought to amend the protective order to restrict AI use with confidential information. The discoverability of the plaintiff’s actual prompts and outputs was not directly at issue, but in working through the questions before it, the court engaged in an analysis that pushes back on several of the premises underlying Heppner—and the result is some useful support for the proposition that AI-assisted litigation materials are protected work product.

First, the court held that Rule 26(b)(3) broadly protects materials prepared in anticipation of litigation by a party, not merely by counsel. In civil cases, the rule’s plain language extends work product protection to parties and their representatives—a point that distinguished Morgan from Heppner, which was a criminal case governed by a different procedural rule.

Second, the court rejected the argument that using a third-party platform destroys confidentiality or waives protection. Drawing on Fourth Amendment case law, the court made a point that is both memorable and important: does anyone with a Gmail account forfeit all rights to confidentiality because Gmail has access to their emails? The court emphasized that in today’s world, nearly all electronic interactions pass through third-party systems, and courts are increasingly pushing back on the notion that using modern technology automatically destroys privilege.

Third, the court emphasized that work product (in contrast to attorney/client privilege) waiver requires disclosure to an adversary or under circumstances that substantially increase the likelihood that an adversary would obtain the information. Disclosure to an AI platform provider does not meet this standard—you are not disclosing information to your adversary, and there is no meaningful indication the information will end up in an adversary’s hands simply because it was entered into an AI system.

An important caveat: because the pro se litigant in Morgan was simultaneously the party and the advocate, Morgan does not resolve what happens when a non-attorney party — say, a corporate employee — uses AI independently, without attorney direction. In that scenario, the Heppner “gap” between party and counsel would exist even in the civil context. But the textual foundation of Rule 26(b)(3), as the Morgan court discussed it, supports an argument that such use would still be protected.

The bottom line is that while Morgan did not directly rule on the discoverability of prompts and outputs, it supports the position that AI-assisted litigation materials can fall within work product — and that the mere fact that information is processed or stored by an AI platform does not automatically result in waiver. This is a practical and sensible position that hopefully other courts will follow.

Tremblay: Prompts and Outputs as Opinion Work Product

As the Blog has noted (here), Tremblay v. OpenAI Inc., No. 23-cv-03223-AMO, 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024), established that when counsel crafts AI prompts, both the prompts and the resulting outputs constitute opinion work product — the highest tier of protection — because the prompts reflect counsel’s mental impressions and opinions about how to interrogate the AI tool. Opinion work product is virtually undiscoverable absent a showing that counsel’s mental impressions are at issue and the need for the material is compelling.

This means that, at least until the case law further develops, having counsel create or direct the creation of AI prompts is not just a best practice. It has doctrinal consequences for the level of protection those materials receive. As a general principle, the closer an attorney is to the creation of the prompts, the stronger the argument that the materials are protected work product.

Waiver: Where Protection Breaks Down

Work product protection can be lost, and several recent decisions illustrate how that can happen in the context of AI use.

In Concord Music Group, Inc. v. Anthropic PBC, the court addressed waiver across multiple discovery orders over the course of 2025. See, e.g., 2025 WL 1482734 (N.D. Cal. May 23, 2025); 2025 WL 3677935 (N.D. Cal. Dec. 18, 2025). Consistent with Tremblay, the court held that the prompts and related settings used in the publishers’ investigations were opinion work product. The publishers voluntarily produced the prompts and outputs they relied on in their complaint and filings — roughly 5,000 prompt-output pairs. The defendant then sought broadly all of the remaining prompts and outputs, including those that did not support the publishers’ claims. The court denied that request as overbroad, holding that waiver must be closely tailored to the needs of the opposing party and limited to what is necessary to rectify any unfair advantage gained.

Later in the litigation, the publishers conducted an investigation into the effectiveness of the defendant’s AI guardrails — essentially testing whether the guardrails could be circumvented using various prompts. When they indicated they intended to put a witness on the stand to testify about that investigation — including the prompts the witness used and what he learned using them — the court found at-issue waiver under the sword-and-shield doctrine. In deciding where the draw the line, the court held that protection had been waived for the prompts that had been provided to the witness to conduct the investigation—including those provided to him by counsel—and the associated outputs, but those that remained with the attorney (and therefore were not placed at issue) remained protected. So here we see waiver of opinion work product, but a narrow tailoring of that waiver by the court.

A similar result played out in T.B. v. Big Brothers Big Sisters of N.Y.C., No. 452864/2021, 2025 WL 2443502 (N.Y. Sup. Ct. Aug. 21, 2025), where plaintiff’s counsel disclosed a portion of a ChatGPT transcript during a deposition. The court held that this placed the summary at issue and waived work product protection, and ordered production of the full AI exchange.

While these waiver principles are not new (or unique to AI), what AI potentially changes is the frequency with which the issue might arise. Tools that make it easy to generate analytical work product—evaluating complaint data, synthesizing the medical literature, identifying patterns in adverse event reports—also make it easy to rely on that work product in ways that place it at issue. The discipline and forethought are the same that have always been required. AI may just create more opportunities to get it wrong.

Where This Is Headed

As Judge Rakoff observed, AI “presents a new frontier in the ongoing dialogue between technology and the law.” That much is obvious. What is increasingly clear from the case law, however, is that courts are finding ways to apply existing doctrines to new technology. Work product protection extends to AI interactions. Attorney-crafted prompts are opinion work product and therefore receive the highest level of protection. And the mere use of a third-party AI platform does not automatically waive that protection.

The attorney/client privilege question is less settled, but the analogies are there. And there is no reason to think that courts will not adapt privilege doctrine to accommodate AI—just as they have adapted it to accommodate other evolutions in the provision of legal services. The waiver landscape requires attention, but the principles exist and the framework for applying them is well established. What remains is for courts to work through the application in this particular context.

Drug & Device Resources

Bexis’ Book

FDA-regulated products now account for an estimated one-fifth of overall economic activity in the U.S. They have also been the focus of a litigation explosion. This timely guide covers all aspects of litigation involving drugs, medical devices, vaccines and other FDA-regulated prescription products.
Learn more about this book