Photo of Bexis

We recently received an email from a long-time reader calling our attention to a litigant’s bizarre artificial-intelligence-aided shenanigans in Conlan Tire Co., LLC v. Gonzales, 2026 WL 676450 (E.D. Mich. March 10, 2026).  This wasn’t your run-of-the-mill AI-generated citations situation, but rather a use of ChatGPT that provided revealing insights into the litigant’s sanctionable intentions.  The litigant in Conlan Tire wasn’t a personal injury plaintiff trying to eliminate adverse evidence – he was a defendant in an unfair competition lawsuit – but he just as well could have been.

In Conlan, our devious litigant denied signing a critical agreement with the plaintiff, his erstwhile employer, and claimed that an image of his signature lifted from something else had been perfidiously added to the agreement.  But digital forensic examination his cellphone exposed an interesting conversation with ChatGPT:

[T]he digital forensic examiners discovered that [the litigant] had used ChatGPT to research how and when deleted emails could be retrieved and produced by Yahoo under a subpoena.  ChatGPT provided the following responses to [the litigant’s] pointed queries:  Yahoo could not retrieve an email deleted from his account a year ago, even under a subpoena; an email permanently deleted today would be retrievable by Yahoo for at most seven days, and Yahoo does not maintain individual deletion histories nor does it track deleted emails in an audit log.  [The litigant] provided the forensic examiner access to his Yahoo account 28 days after he conducted his ChatGPT research on deleted Yahoo emails.

2026 WL 676450, at *2 (citations and footnote omitted).  See also Conlan Tire Co., LLC v. Gonzales, 2025 WL 4704197, at *2 (E.D. Mich. Aug. 11, 2025) (discussing the same facts in greater detail). 

This litigant’s use of ChatGPT supported Conlan’s grant of spoliation sanctions and rejection of the litigant’s “contrived, implausible, and deceptive arguments that his ChatGPT queries were not indicative of his intent to delete emails.”  Id. at *3.  The adverse inferences in large part led the court to grant the plaintiff’s requested injunction against the defendant.  2025 WL 4704197, at *5.

This is not the first time the Blog has chronicled litigants (or their lawyers) using artificial intelligence to do stupid things.  We’ve discussed litigants using AI to do their own legal research unbeknownst even to their lawyers.  And like many others, we’ve bewailed the plague of “hallucinated” AI-generated fake citations in legal papers and even expert reports.

As a result, we’ve decided to review every judicial decision that mentions “ChatGPT” (well over 500 decisions) and see how that system – the first and foremost generally available generative AI large language model – has been used and misused in actual litigation.  We’re not trying to pick on ChatGPT.  Other LLMs share these risks.  However, ChatGPT is both the longest existing such platform, and often cited as the prime example of this type of AI (we found lots of “such as ChatGPT” references that we are not including here).  Another reason for focusing on ChatGPT is that it was not designed for legal work and “is generally understood to be the most problematic AI platform to use for drafting legal filings.”  In re Hines, 2025 WL 3137723, at *7 (Bankr. S.D. Ohio Nov. 7, 2025).

The majority of the decisions that mention ChatGPT substantively (others reference it in facts not having legal significance, and still others involve suits directly against ChatGPT’s owners) involve judicial orders of two types:  either (1) orders requiring litigants to disclose use of AI generally, with ChatGPT being one of list of named AI sites; or (2) orders forbidding litigants from submitting confidential discovery materials to AI sites that are not totally closed-source or proprietary to recipient counsel, again with ChatGPT again being one of a list.  We’re not including those decisions, but they’re out there in abundance if anybody needs them. We’re not including those decisions, but they’re out there in abundance if anybody needs them.  One recent decision, however, refused to modify an existing protective order to allow a pro se plaintiff to upload confidential information into ChatGPT:

Plaintiff believes the Court should permit him to use ChatGPT to review confidential evidence because he is pro se and does not have support. . . .  Defendant argues that, aside from being unnecessary, there is substantial risk in using a public large-language model platform to process confidential information. Defendant represents that uploading confidential evidence to ChatGPT risks disclosure of [what] . . . Plaintiff acknowledges to be “sensitive information.”

The Court finds that the balance of conflicting interests weighs in favor of Defendant’s position.  It is not disputed that exposing confidential information to ChatGPT creates at least some risks with respect to how that information will be used and safeguarded.

Baqleh v. Altamirano, 2026 WL 3257576, at *2 (N.D. Cal. Oct. 28, 2026) (citations omitted).

Only a minority of the decisions mentioning ChatGPT are of the same sort of interest as Conlon.  But because they are the most relevant, both to us and to our readers, we’re starting with them.  Some of the cases involving ChatGPT concern discovery, which we expect will be a recurring issue going forward.  In Villanueva v. Las Vegas Metropolitan Police Dept., 2026 WL 2942348, at *2 (D. Nev. Sept. 29, 2026), the court ordered production of a plaintiff’s expert’s ChatGPT reliance after plaintiff failed to disclose the expert’s AI use prior to the expert’s deposition.  Id. at *3. However, the expert’s “questions and prompts to ChatGPT” were protected work product. Id. at *3-4.  Further, the expert’s “creation of a draft report on ChatGPT is protected as a draft report, because . . . some of this information was directly incorporated into [the expert’s] own written report.”  Id. at *4. 

Concord Music Group, Inc. v. Anthropic PBC, 2025 WL 1482734 (N.D. Cal. May 23, 2025), held that “Chat GPT prompts” run by lawyers during “pre-suit investigation” were protected work product.  Id. at *2 (“failed prompts and related settings are attorney work product”).  Concord Music followed Tremblay v. OpenAI, Inc., 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024), which likewise held that certain “ChatGPT prompts were queries crafted by counsel and contain counsel’s mental impressions and opinions about how to interrogate ChatGPT” about issues relevant to the litigation.  Id. at *2 (citation omitted).

In Jones v. Delta Air Lines, Inc., 2026 WL 1091186, at *4-5 (E.D. Mich. April 22, 2026), a pro se plaintiff’s attorney/client privilege claim that ChatGPT was acting as her attorney was rejected, and plaintiff was directed to answer questions about her use of the service.  Accord, Lucas v. Scharf, 2026 WL 642917, at *2 n.2 (W.D.N.C. Feb. 25, 2026) (“‘conversations’ with AI tools such as ChatGPT are not privileged and may be subject to discovery”).  On the other hand, Warner v. Gilbarco, Inc., 820 F. Supp.3d 629, 636-37 (E.D. Mich. 2026), held that “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background,” and thus a pro se plaintiff’s interactions with the program received work product protection.

Even more seriously, Flexport, Inc. v. Freightmate AI, Inc., 2026 WL 2771671 (N.D. Cal. Sept. 17, 2026), imposed spoliation sanctions under Fed. R. Civ. P. 37(e) for a litigant’s destruction of several files containing “prompts used with ChatGPT.”  Id. at *8 (ordering “for the jury to be instructed on the law on suppression of evidence” to “determine whether or not to consider the lost ESI would have been unfavorable to Defendants”).

As we have discussed previously, another group of cases has excluded expert witnesses because the presence of ChatGPT-generated hallucinations in their reports “shattered” their “credibility.”  Ledoux v. Outliers, Inc., 2026 WL 2408808, at *7 (W.D. Wash. Aug. 18, 2026); Kohls v. Ellison, ___ F. Supp.3d ___, 2025 WL 66514, at *4 (D. Minn. Jan. 10, 2025).  Recently, an unpublished (and thus not citable in the state) California state appellate decision decided with respect to ChatGPT generated expert opinions:

We agree with [petitioner] in principle that the use of AI by a forensic psychologist to generate an expert opinion, at present, would constitute a novel scientific technique.  Any such opinion, therefore, would be inadmissible absent an adequate foundation demonstrating the reliability of the psychologist’s methodology, including the general acceptance of the new technique in the relevant scientific community.

Gillespie v. Superior Court, 2026 WL 3333170, at *8 (Cal. App. Nov. 4, 2026) (citations and quotation marks omitted).

Other forms of ChatGPT-generated “evidence” have also been excluded.  Buehler v. Nationstar Mortage [sic] LLC, 2026 WL 2349853, at *2 (W.D. Wash. Aug. 13, 2026), excluded Chat GPR-generated “calculations” where the proffer included “no explanation as to what information was given to ChatGPT, who provided such information, or under what circumstances the Rules of Evidence would allow the Court to accept ‘simple calculations’ from ChatGPT to evaluate the merits of [plaintiff’s] claim.”  Lepp v. Mallett, 2026 WL 125968, at *6 (E.D. Mich. Jan. 16, 2026), refused to consider a pro se plaintiff’s ChatGPT-generated “legal analysis report” as improper (“the place for legal arguments is in the brief, not the exhibits”) and unreliable since ChatGPT “do[es] not − and cannot − decide what is true and what is false.”  See also Veleber v. Alphabet, Inc., 2026 WL 2475778, at *1-2 (N.D. Cal. Aug. 24, 2026) (holding that plaintiff’s “purported ‘declarations’ or ‘statements’ from AI systems” (including ChatGPT) were “fanciful allegations” and dismissing case); Wilson v. KIPP Texas, Inc., 2025 WL 3030444, at *1, 3 (N.D. Tex. Oct. 29, 2025) (sanctioning counsel for submitting as evidence “11 non-existent statements” generated by ChatGPT); United States v. Sullivan, 2025 WL 2402894, at*1 (D. Haw. Aug. 19, 2025) (ChatGPT-generated “data . . . was inaccurate and misleading, rendering the same useless for any purpose”); Pop Top Corp. v. Rakuten Kobo, Inc., 2025 WL 2098597, at *4 (N.D. Cal. July 25, 2025) (bankruptcy debtor’s “proffered ChatGPT analysis [of his deposition] lacks any independent corroboration and thus is not reliable”).  But see Ferlito v. Harbor Freight Tools USA, Inc., 2025 WL 1181699, at *4 (E.D.N.Y. April 23, 2025) (accepting expert’s testimony that he was “quite amazed” that his allegedly subsequent “ChatGPT search confirmed what he had already opined” and finding “no indication” what the expert had relied on “ChatGPT to generate a report with false authority or that his use of AI would render his testimony less reliable”).

Relevant to the consideration of a possible Fed. R. Evid. 707, the use of ChatGPT-generated information as evidence, without supporting expert explication, has been rejected.  In Pegnatori v. Pure Sports Technologies LLC, 2023 WL 6626159 (D.S.C. Oct. 11, 2023), a patent case, the court refused to credit a ChatGPT query as to “the definition of” a disputed technical term (“foam”).  Id. at *5.  “[T]he court would be taking its eye off the ball if it applied the ChatGPT definition in its review of extrinsic evidence.”  Id. at *6 (citation omitted).

Several cases, all involving the same law firm, have repeatedly refused to consider ChatGPT generated information purporting to opine on the reasonableness of fee demands.  Reliance on AI, with no accompanying expert evidence, has universally been rejected in this context.

[I]nvocation of ChatGPT as support for its aggressive fee bid is utterly and unusually unpersuasive.  As the firm should have appreciated, treating ChatGPT’s conclusions as a useful gauge of the reasonable billing rate for the work of a lawyer with a particular background carrying out a bespoke assignment for a client in a niche practice area was misbegotten at the jump. . . .  In claiming here that ChatGPT supports the fee award it urges, the [firm] does not identify the inputs on which ChatGPT relied.  It does not reveal whether any of these were similarly imaginary.  It does not reveal whether ChatGPT anywhere considered a very real and relevant data point:  the uniform bloc of precedent, canvassed below, in which courts in this District and Circuit have rejected as excessive the [firm’s] billing rates. . . .  The Court therefore rejects out of hand ChatGPT’s conclusions as to the appropriate billing rates here.  Barring a paradigm shift in the reliability of this tool, the [firm] is well advised to excise references to ChatGPT from future fee applications.

J.G. v. New York City Dept. of Education, 719 F. Supp.3d 293, 30-08 (S.D.N.Y. 2024) (citations omitted).  Unfortunately, that firm did not take that advice, and continued submitting such “evidence”:

Additional sources provided by [the firm] are also unpersuasive, and I do not recommend placing substantial weight on any of them.  Specifically, [it]references the artificial intelligence tool “ChatGPT.”  [It] relies on ChatGPT’s feedback to demonstrate what a client’s search may provide when attempting to determine hourly rates for IDEA litigation.  Here, [this] reliance on ChatGPT is inappropriate, because ChatGPT has been shown to be an unreliable resource.

D.S. v. New York City Dept. of Education, 2024 WL 2159785, at *6 (Mag. S.D.N.Y. April 29, 2024) (citations omitted), adopted, 2024 WL 2158583 (S.D.N.Y. May 14, 2024).  See also S.M. v. New York City Dept. of Education, 2025 WL 4050877, at *10 (Mag. S.D.N.Y. Nov. 10, 2025) (same as J.G.), adopted, 2026 WL 104183 (S.D.N.Y. Jan. 14, 2026); K.K. v. New York City Dept. of Education, 2024 WL 4203783, at *6 n.11 (S.D.N.Y. Aug. 22, 2024) (“invocation of ChatGPT as support for [an] aggressive fee bid is utterly and unusually unpersuasive”), adopted, 2024 WL 4203251 (S.D.N.Y. Sept. 16, 2024); S.C. v. New York City Dept. of Education, 2024 WL 3518522, at *3 (S.D.N.Y. July 24, 2024) (“reject[ing] results from ChatGPT-4 as nonprobative and unreliable” on fee calculations); Z.H. v. New York City Dept. of Education, 2024 WL 3385690, at *5 (S.D.N.Y. July 12, 2024) (same as J.G.); R.R. v. New York City Dept. of Education, 2024 WL 6046877, at *6 (Mag. S.D.N.Y. June 3, 2024) (same as D.S.), adopted, 2025 WL 3516102 (S.D.N.Y. Dec. 8, 2025); M.B. v. New York City Dept. of Education, 2024 WL 1435330, at *4 (Mag. S.D.N.Y. Jan. 24, 2024) (“declin[ing] to consider ChatGPT’s claims” since it is “notorious for inventing false information”), adopted, 2024 WL 1343596 (S.D.N.Y. March 30, 2024).

Nor can “conversation with ChatGPT” be considered as “newly discovered evidence.”  United States v. Chin, 2024 WL 251496, at *1 (D.C. Cir. Jan. 23, 2024).  United States v. Jones, 2026 WL 615699 (S.D. Ohio March 5, 2026), held that a losing party cannot create newly discovered evidence belatedly “from a question put to Chat GPT.”  “While the chatbot is not licensed to practice law and it was careful to distinguish what it reported from legal advice, its output cannot be properly classified as making Defendant’s claim newly ripe.”  Id. at *2.  Accord United States v. Bauer, 2025 WL 2084917, at *3 (N.D. Cal. July 24, 2025) (same result).

If you think that AI has created a license for lazy lawyers and lazier litigants to run amok, you would be right.  The bulk of Chat-GPT-related decisions that that we found concern case-specific (ab)use of ChatGPT (similarly to what can happen with other AI) involve its propensity to generate hallucinated citations, quotations, etc., in legal papers.  We start with the appellate caselaw.  “The use of AI is now so pervasive in legal practice that attorneys can no longer credibly claim ignorance of its pitfalls, including its propensity to hallucinate legal authority.”  Douglas v. Deutsche Bank National Trust Co., 363 A.3d 1088 (D.C. 2026).  “[T]he use of artificial intelligence must be accompanied by the application of actual intelligence in its execution.”  Id. at 1091 (citation and quotation marks omitted). 

Today we join the chorus reminding attorneys while generative AI may be a useful tool, it is no substitute for human intellect and professional judgment.  The duty to verify the law before presenting it to a court is woven into the profession itself, and while technology may evolve at breathtaking speed, the ethical obligations of lawyers cannot bend to the speed of progress.  A lawyer’s duties cannot be delegated to a machine capable of confidently inventing what it cannot legitimately find.

Southland Homes & Real Estate & Investment, LLC v. Lam, 348 Cal. Rptr.3d 473, 475 (Cal. App. 2026) (sanctioning ChatGPT-generated citations in appellate briefing).  See Perez-Castillo v. Blanche, 177 F.4th 837, 845-46 (7th Cir. 2026) (attorney sanctioned for claiming that “ChatGPT ‘hallucinate[d] facts not present in the record in this matter and insert[ed] them into the draft petition” and that it made her “correct case citations incorrect”); Park v. Kim, 91 F.4th 610, 615 (2d Cir. 2024) (arguments based on “non-existent precedent generated by ChatGPT” are an “abuse of the adversary system”) (citation and quotation marks omitted); Tov Realty, LLC v. Suarez, 360 A.3d 1062, 1066 (Conn. 2026) (attorney claiming that ChatGPT surreptitiously added hallucinated citations to a properly prepared appellate brief sanctioned); State v. Sandoval, 2026 WL 2661673, at *1-2 (N.M. Sept. 9, 2026) (attorney held in “direct contempt” and barred from any future appearances for submitting ChatGPT-generated “factual and legal misrepresentations,” including “wholly fabricated witnesses”); State ex rel. Oklahoma Bar Ass’n v. Reeves, ___ P.3d ___, 2026 WL 1480563, at *2 (Okla. May 27, 2026) (attorney publicly reprimanded for repeatedly submitting papers with hallucinated ChatGPT citations that “were completely made up”); Ex parte Thomas, ___ S.W.3d ___, 2026 WL 2880076, at *8 (Tex. App. Sept. 24, 2026) (dismissing appeal as inadequately briefed due to multiple ChatGPT hallucinated citations); Scott v. Illinois Human Rights Commission, ___ N.E.3d ___, 2026 WL 2167461, at *6-8 (Ill. App. July 28, 2026) (attorney sanctioned; after filing two ChatGPT generated briefs, the second a response to the court’s order regarding hallucinations in the first brief; “no matter how much one pays for ‘premier’ or ‘corporate’ versions of AI products, it does not negate an attorney’s obligation to verify all citations of authority”); State v. Coleman, 280 N.E.3d 1042, 1072 (Ohio App. 2026) (attorney who repeatedly filed ChatGPT hallucinated papers held in contempt, disqualified, fined, and more), appeal allowed, 282 N.E.3d 834 (Ohio 2026); Matter of Zareh, 250 N.Y.S.3d 534, 538 (N.Y.A.D. 2026) (attorney publicly censured for ChatGPT hallucinations); Mezu v. Mezu, 346 A.3d 181, 188, 190 193 (Md. App. 2025) (referring counsel for discipline for including ChatGPT “citation irregularities” in appellate briefing; “[a] competent attorney reads the legal authority cited in court pleadings”); In re Kenney, 422 So.3d 905, 927 (La. App. 2025) (sanctions affirmed against ChatGPT-reliant attorney who cited “fabricated” cases); Noland v. Land of the Free, L.P., 336 Cal. Rptr.3d 897, 913-14 (Cal. App. 2025) (pervasive hallucinations in ChatGPT-using lawyer’s brief justified declaring the appeal “frivolous”); Garner v. Kadince, Inc., 571 P.3d 812, 815 (Utah App. 2025) (“every attorney has an ongoing duty to review and ensure the accuracy of their court filings” and counsel here “fell short of their gatekeeping responsibilities . . . when they submitted a petition that contained fake precedent generated by ChatGPT”; monetary sanctions imposed).

The most frequently cited trial court decision addressing hallucinated ChatGPT material in attorney-filed legal papers is Mata v. Avianca, Inc., 678 F. Supp.3d 443 (S.D.N.Y. 2023), in which an attorney “submitted non-existent judicial opinions with fake quotes and citations created by the tool ChatGPT, then continued to stand by the fake opinions after judicial orders called their existence into question.”  Id. at 448.

Many harms flow from the submission of fake opinions.  The opposing party wastes time and money in exposing the deception.  The Court’s time is taken from other important endeavors.  The client may be deprived of arguments based on authentic judicial precedents.  There is potential harm to the reputation of judges and courts whose names are falsely invoked as authors of the bogus opinions and to the reputation of a party attributed with fictional conduct.  It promotes cynicism about the legal profession and the American judicial system.

Id. at 448 (footnote omitted).  Thus, Mata held that “[a]n attempt to persuade a court or oppose an adversary by relying on fake opinions is an abuse of the adversary system.”  Id. at 461 (citation omitted).  Both the offending lawyer and his law firm were sanctioned under Rule 11 and the court’s inherent powers upon the court finding subjective bad faith.  Id. at 464-65.

Other trial court opinions addressing ChatGPT-generated hallucinations are: Miller v. Regions Bank, 833 F. Supp.3d 1201, 1213, 1215, 1218-19, 1221 (N.D. Ala. 2026) (attorney held in civil contempt and suspended from practicing before the court for filing a brief with numerous ChatGPT-generated hallucinations, “obfuscating” his use, and then deleting his account after being ordered to produce it to the court); Rivera v. Triad Properties Corp., 829 F. Supp.3d 983, 1029 (N.D. Ala. 2026) (disqualification, monetary, and other sanctions imposed on an attorney who “pair[ed] his [ChatGPT] misuse with a refusal to accept responsibility and further false statements to the court”); State National Insurance Co., Inc. v. Treadwell, 826 F. Supp.3d 1352, 1357 (N.D. Ala. 2026) (attorney publicly reprimanded for submitting legal papers containing ChatGPT hallucinated citations); Barkley v. City of Tempe, 2026 WL 3015647, at *4 (D. Ariz. Oct. 7, 2026) (parties filing “paper[s] incorporating inaccurate or undeclared generative AI outputs, including but not limited to inaccurate or non-existent case citations” subject to sanctions “without further warning”); Tiekert v. Village of Mamaroneck, 2026 WL 2690037, at *6-8 (S.D.N.Y. Sept. 14, 2026) (attorney held in bad faith and sanctioned for using ChatGPT generated material “only the most superficial level of checks”; attorney used ChatGPT “without verification of its output” despite awareness of “the risk of hallucination”); Beus Gilbert PLLC v. Brigham Young University, 2026 WL 2654907, at *1 (D. Utah Sept. 9, 2026) (monetary and other sanctions imposed on lawyers who submitted ChatGPT-generated hallucinations in their briefing); Ruiz v. Magellan Financial & Insurance Services, 2026 WL 2469948, at *2-3 (D. Ariz. Aug. 24, 2026) (given “the presence of three aggravating factors (prior discipline, a pattern of misconduct/multiple offenses, and substantial experience),” lawyer received formal public reprimand for repeated submission of papers with ChatGPT-generated hallucinations); DeFord v. Koutoulas, 2026 WL 2569638, at *9 (M.D. Fla. Aug. 4, 2026 (ChatGPT using counsel and client suffered default as a sanction for, inter alia “submit[ing] fabricated authority across at least eight separate filings”); Smith v. Verra Mobility Corp., 2026 WL 1197241, at *1 (M.D. Fla. May 1, 2026) (attorney sanctioned for “negligently failing to verify multiple authorities cited [by ChatGPT and other tools] in his motion”); Gentry v. Thompson, 2026 WL 787563, at *2 (E.D. La. March 20, 2026) (“Using ChatGPT and failing to check the case citations it provides constitute not only a lapse in professional judgment, but also a violation of [Rule 11]”; sanctions imposed); Kattom v. Bondi, 2026 WL 637419, at *1-2 (W.D. La. March 6, 2026) (attorney fined for ChatGPT hallucinated citations); Allah-El v. Columbia University, 2026 WL 412646, at *1, 3-4 (S.D.N.Y. Feb. 13, 2026) (attorney who abused ChatGPT denied pro hac vice admission); Lexos Media IP, LLC v. Overstock.com, Inc., 2026 WL 265581, at *15-17 (D. Kan. Feb. 2, 2026) (attorney directly responsible ChatGPT-generated hallucinations fined $5000 and had pro hac vice revoked; other attorneys received lesser sanctions); Mattox v. Product Innovations Research, LLC, 807 F. Supp.3d 1341, 1347-48, 1350-52 (E.D. Okla. 2025) (multiple counsel sanctioned for submitting eleven pleadings with ChatGPT-hallucinated citations; “The volume and repetition of these errors show not inadvertence but systemic neglect.”; all implicated pleadings stricken); United States v. McGee, 806 F. Supp.3d 1264, 1274-75 (S.D. Ala. 2025) (“insertion of bogus citations is not a mere typographical error, nor the subject of reasonable debate. It is just wrong.”; publicly funded criminal defense lawyer must reimburse the fund, among other sanctions); ByoPlanet International, LLC v. Johansson, 792 F. Supp.3d 1341, 1351 (S.D. Fla. 2025) (multiple filings with ChatGPT-generated hallucinations “constitutes repeated, abusive, bad-faith conduct that cannot be recognized as legitimate legal practice and must be deterred”); Johnson v. Dunn, 792 F. Supp.3d 1241, 1266-67 (N.D. Ala. 2025) (law firm partners with “access to gold-standard legal research databases” who nonetheless filed documents with ChatGPT generated hallucinations were publicly reprimanded, disqualified, and referred to disciplinary authorities); N.Z. v. Fenix International Ltd., 2025 WL 3626155, at *2, 4-5 (C.D. Cal. Dec. 12, 2025) (mass tort lawyers and their local counsel “fail[ed] to verify the validity” of ChatGPT-generated material; monetary sanctions and sending copy of order to “each plaintiff” imposed); Hill v. Oklahoma County Criminal Justice Authority, 2025 WL 4036934, at *3-4 (Mag. W.D. Okla. Sept. 5, 2025) (filing of multiple briefs with ChatGPT hallucinations resulted in monetary and non-monetary sanctions), adopted, 2025 WL 3490507, at *2 (W.D. Okla. Dec. 4, 2025); Coronavirus Reporter Corp. v. Apple, Inc., 2025 WL 2162947, at *5-6 (N.D. Cal. July 30, 2025) (payment of counsel fees imposed as a sanction against attorneys who “used ChatGPT to generate in its entirety an exhibit” that they submitted on three separate occasions); Nextpulse, LLC v. Life Fitness, LLC, 2025 WL 2453720, at *3-4 (N.D. Ill. Mar. 7, 2025) (submission of “fraudulent case law” generated by ChatGPT, followed by “general lack of candor” by the attorney resulted in monetary sanctions and referral to disciplinary board); Bunce v. Visual Technology Innovations, Inc., 2025 WL 662398, at *3 (E.D. Pa. Feb. 27, 2025) (sanctioned counsel had “blindly trusted an algorithm he had never used before” and “conducted no research into ChatGPT’s efficacy as a legal tool”); Iovino v. Michael Stapleton Associates, Ltd., 2024 WL 3520170, at *7 (W.D. Va. July 24, 2024) (sanctions order to show cause where hallucinated case law was “the result of ChatGPT run amok”) (citation and quotation marks omitted); Pegnatori v. Pure Sports Technologies, LLC, 2023 WL 6626159, at *5 (D.S.C. Oct. 11, 2023) (“thus far ChatGPT’s batting average in legal briefs leaves something to be desired”; “attorneys have faced admonishments and sanctions from courts in which they made these improper and filings.”); In re Martin, 670 B.R. 636, 649 (Bankr. N.D. Ill. 2025) (counsel’s “alleged lack of knowledge of ChatGPT’s shortcomings leads me to do what courts have been doing with increasing frequency: announce loudly and clearly (so that everyone hears and understands) that lawyers blindly relying on generative AI and citing fake cases . . . will be sanctioned”); People v. Crabill, 2023 WL 8111898, at *1 (Colo. O.P.D.J. Nov. 22, 2023) (public reprimand of attorney for ChatGPT-generated hallucinations “falsely attributed . . . to a legal intern”). Cf. Walters v. Openai, L.L.C., 2025 WL 2979163, at *8-9 (Ga. Super. May 19, 2025) (summary judgment granted in defamation action because ChatGPT’s false answers to prompts could not have been made with “actual malice”; the program’s proclivity to hallucinate was both well-known and warned about).  Contra Dodge v. FirstService Residential Arizona LLC, 2025 WL 3653164, at *3 (D. Ariz. Dec. 17, 2025) (“public humiliation . . . is a sufficient remedy” for solo practitioner’s Chat-GPT hallucinations, since he lost the case anyway); Perez v. Evans, 2025 WL 2726792, at *4 (S.D.N.Y. Sept. 25, 2025) (“declining] to impose sanctions on Plaintiff for his violation of Rule 11(b), in part due to Plaintiff’s admission to using ChatGPT”)

Finally, another all-too-common use of ChatGPT is by pro se litigants, who also fail to check the validity of the output, but with arguably more of an ignorance-based excuse.  See Amarsingh v. Frontier Airlines, Inc., 2026 WL 352016, at *7 (10th Cir. Feb. 9, 2026) (pro se appeal held “frivolous as argued” due to reliance on several ChatGPT hallucinated citations); Johnson v. Johnson, 2025 WL 3034722, at *3 n.4 (Ind. App. Oct. 30, 2025) (in table at 270 N.E.3d 542) (criticizing pro se parties on both sides for submission of ChatGPT-generated hallucinations that “severely hindered” appellate review); Woodard v. Southern Trust Insurance Co., 2026 WL 3078487, at *2 (M.D. Ga. Oct. 14, 2026) (requiring signed statements by pro se litigants disclosing any AI use and swearing to the accuracy of “all facts and legal authorities”); George v. Rollins, 2026 WL 3015346, at *1 (D. Ariz. Oct. 7, 2026) (dismissing AI generated pro se complaint as not a “short and plain” statement, and prohibiting plaintiff from using AI to file an amended complaint); Garcia v. City of Monte Vista, 2026 WL 2921068, at *2 (D. Colo. Sept. 29, 2026) (attorney who filed document with false citations with a certification requirement regarding AI usage for all future filings); Butler v. The Travelers Indemnity Co., 2026 WL 3059644, at *5-6 (Conn. Super. Oct. 7, 2026) (striking pro se pleadings containing fake citations and quotations); Johnson v. Nationstar Mortgage LLC, 2026 WL 2516390, at *17-18 (D. Md. Aug. 26, 2026) (dismissing claims of pro se plaintiff who persisted in submitting papers with ChatGPT-generated hallucinated citations despite the court’s “admonition”); Booker v. U.S. Bank National Ass’n, ___ F. Supp.3d ___, 2026 WL 2254689, at *5 (D. Conn. Aug. 5, 2026) (pro se plaintiff sanctioned for ChatGPT hallucinations after “acknowledg[ing] that he had only ‘briefly read’ some of the cited cases, had not read others, and ultimately could not say whether he had read any of them before filing”); Barteca Holdings LLC v. Tacobarn Newtown LLC, ___ F. Supp.3d ___, 2026 WL 2253785, at *7 (D. Conn. Aug. 4, 2026) (attorney sanctioned for ChatGPT use resulting in “submission of unverified AI-generated legal authorities”); McQueen v. Federal National Mortgage Ass’n, 2026 WL 2210113, at *6 (C.D. Cal. July 24, 2026) (ChatGPT-generated pro se complaint dismissed; it “contains circuitous factual assertions, legal commentary, confounding headers, . . . other unnecessary information that renders each paragraph difficult to understand if not incomprehensible,” and “baseless legal analysis and conclusions of law”); Peters v. Banner Health, 2026 WL 143152, at *4-6 (Mag. S.D. Tex. Jan. 15, 2026) (pro se plaintiff’s repeated filings rife with ChatGPT-generated hallucinations was contumacious and warranted dismissal with prejudice), adopted, 2026 WL 157685 (S.D. Tex. Jan. 20, 2026), aff’d, 2026 WL 1611868 (5th Cir. June 4, 2026); Jarrus v. Governor of Michigan, 2025 WL 3461540, at *3 (E.D. Mich. Dec. 2, 2025), reconsideration denied, 2026 WL 27814, at *4 (E.D. Mich. Jan. 5, 2026) (pro se plaintiff sanctioned after being “warned that financial sanctions would be incurred at $200 per violation” –for each ChatGPT-generated hallucination; comparing amount to plaintiff’s annual ChatGPT subscription); United States v. Bendy, 2025 WL 3173897, at *2 (E.D. Tex. Nov. 13, 2025) (court refuses to accept pro se plaintiff’s unverified ChatGPT-generated filings); Morgan v. Sprouts Farmers Market, 2025 WL 2959489, at *1 (D. Ariz. Oct. 20, 2025) (pro se plaintiff who filed “lengthy” complaint including ChatGPT-generated “legal analysis” forbidden from using “Chat GPT or any similar form of artificial intelligence” in any refiled complaint); Powhatan County School Board v. Skinger, 2025 WL 1559593, at *11 (E.D. Va. June 2, 2025) (striking pleadings containing ChatGPT filed by likely judgment-proof pro se plaintiff); Gabriel v. SLS Lux Brickell Hotel, 2025 WL 2020154, at *5 (S.D. Fla. Jan. 15, 2025), (pro se plaintiff whose entirely generated complaint was dismissed “may not use ChatGPT to write his amended complaint”); Dukuray v. Experian Information Solutions, 2024 WL 3812259, at *11 (Mag. S.D.N.Y. July 26, 2024) (while “it is possible Plaintiff is not aware of the risk that ChatGPT and similar AI programs are capable of generating fake case citations and other misstatements of law . . ., it is no more acceptable for a pro se litigant to submit briefs with fake case citations than it is for a lawyer to do so”), adopted, 2024 WL 3936347 (S.D.N.Y. Aug. 26, 2024); Neal v. Sage Software LLC, 2024 WL 6915606, at *2, 7-8 (N.D. Ga. May 3, 2024) (entirely ChatGPT generated pro se complaint dismissed).

The continued employment of ChatGPT by lawyers, in particular, is distressing.  Indeed, the sheer number of cases we found from 2026 suggests that the problem is getting worse, not better.  Thus, we reiterate our prior advocacy of a stronger sanctions regime – particularly given the increased availability of generative AI models specifically designed for legal work.  We also suggest amending Rule 11 to remove AI-generated hallucinations from the requirement of filing a separate sanctions motion, since such violations are inevitably open-and-shut.

Photo of Eric Hudson

Last week we participated in a panel addressing judicial scrutiny and transparency in class action settlements. While our regular drug and device litigation is typically mass torts rather than class actions, a couple of the topics seemed blogworthy.

One of our fellow panelists was a federal judge with a growing reputation for casting sunshine on third-party litigation funding (“TPLF”) and private equity’s involvement in class actions and mass torts. We’ve blogged repeatedly about TPLF (for example, here, here, and here), and the effort for adoption of a uniform federal rule requiring TPLF disclosure continues (most recently, 214 companies submitted a joint letter to the Advisory Committee requesting the adoption of a uniform Federal Rule of Civil Procedure requiring the disclosure of third-party litigation funding agreements – the press release from LCJ is here).

But the involvement of private equity and investment in our country’s litigation system goes much deeper than traditional TPLF.  One example is the current MDL involving claims against banks and settlement administrators alleging that certain settlement administrators (some of which are owned or controlled by private equity) worked with settlement banks to receive millions of dollars in undisclosed amounts. Essentially, kickbacks. See Consolidated Class Action Complaint, In re Class Action Settlement Administration Litigation, MDL No. 3162 (D.D.C. May 18, 2026), ECF No. 123. The allegations in that MDL were a significant topic of discussion by the panel, and that spun off some tangential discussion about the impacts of Arizona’s Alternative Business Structure (“ABS”) rule. If you’re not familiar with the ABS model, it permits non-lawyer ownership of law firms and generated a rush by mass tort firms to establish firms in which private equity interests could own a stake in a law firm. A good summary of some of the problems associated with ABS is here, and the directory of entities registered as ABSs in Arizona is here. Indeed, even a publicly traded Australian plaintiff’s class action firm (Shine) registered to do business in the United States as an ABS. The American Bar Association’s Model Rule 5.4—which prohibits attorneys from providing services through a professional corporation or association in which a nonlawyer either owns an interest or has a right to control the lawyer’s professional judgment—precluded non-lawyer ownership of law firms. But the ABS structure in Arizona, which ostensibly was intended to improve access to justice in the state, explicitly removes that prohibition and opens the gates to non-lawyer ownership.

Continue Reading TPLF, ABS and Vertical Integration of Mass Torts Infrastructure

These days, even the ordinary lay press writes a fair amount about what is, and is not, good science.  You would have to have a worm in your brain to miss the debates over whether health effects of this or that substance is supported by good and credible scientific studies or junk science, whether it be vaccines or peptides or beef tallow or whatever.

And there is a lot of bad science out there.  AI fabrications and predatory publishers mean that citations to “published” study papers in “peer reviewed” “journals” should be deemed suspect until vetted by someone with a wary nature and the scientific chops to ask probing questions.

Discerning good science from bad is, of course, part of our everyday Rule 702 practice on the product liability side of the life sciences house. 

But solid clinical trials are even more important on the research and development side of the house.  So our readers may be interested to know that the FDA has issued a revised draft guidance indicating that in some circumstances, just one well-done clinical trial may be sufficient for a manufacturer to demonstrate the “substantial evidence of effectiveness” necessary for approval of a new drug application (NDA), biologics license application (BLA), or supplement.

That’s a big change. The new draft guidance document is Demonstrating Substantial Evidence of Effectiveness for Human Drug and Biological Products (June 2026), and it is intended to replace a 1998 guidance as well as a December 2019 version of this draft guidance.

The guiding light of this guidance is, of course, the statutory definition of “substantial evidence” of drug efficacy from FD&C Act § 505(d)/21 U.S.C. § 355(d):  That is, evidence consisting of adequate and well‑controlled investigations by qualified experts, sufficient to support a fair and responsible conclusion that the drug will have the represented effect under labeled conditions of use. 

Key Factors Affecting the Strength of Clinical Trial Evidence

The guidance addresses five factors for testing the strength of a clinical trial under the “substantial evidence” inquiry:

1.         Trial design.   The trial must be “adequate and well‑controlled” in the sense of 21 C.F.R.  § 314.126—i.e., capable of isolating the drug effect from other influences (spontaneous disease fluctuation, placebo effect, biased observation).  The guidance discusses methods for control and randomization, how non-inferiority (NI) trials can be acceptably designed, and numerous other design topics, from blinding to endpoints to clinical relevance.

2.         Trial conduct.  The clinical trial must not only be well designed, it must also be well run.  FDA will be looking at Good Clinical Practice compliance, data quality, protocol compliance, management of missing data, participant attrition, protection of interim trial results while the trial is ongoing, and the like.

3.         Trial analysis plan.  The right statistical considerations must be evaluated and in place, and if the prior evidence of effectiveness is weak and the sponsor hopes to rely on just one study, then more stringent significance levels will be required.  An adequate and well-controlled trial needs a pre-specified primary analysis aligned with a clinically meaningful estimand (the specific target quantity or treatment effect intended to be measured).

4.         Trial results. The trial results must be clinically meaningful, not just statistically significant.  The persuasiveness of the study results will depend on:

•           The magnitude and clinical relevance of the observed effect;

•           The associated uncertainty;

•           The degree of consistency across multiple relevant endpoints; and

•           Robustness to plausible violations of analytic assumptions.

5.         The overall development program.  A single efficacy study also will be evaluated in the broader context of the drug’s development, such as

•           Early‑phase data on mechanism, dose selection, and related indications;

•           External evidence on disease pathophysiology and natural history;

•           Evidence from related diseases or drugs with similar mechanisms.

Three Pathways to Demonstrate Substantial Evidence of Efficacy

Having laid out expectations for evaluating efficacy trials, the guidance identifies three pathways for using quality clinical trials to meet the substantial evidence of efficacy (even as it acknowledges that these three pathways are not exclusive):

A. One adequate and well‑controlled clinical investigation plus confirmatory evidence

FDA devotes substantial attention to circumstances under which a single adequate and well‑controlled clinical investigation, combined with confirmatory evidence, may suffice.

The Agency indicates that it will consider:

•           Trial design, conduct, analysis, and persuasiveness of the single trial;

•           Source and strength of confirmatory evidence;

•           Disease characteristics (seriousness, unmet need, prevalence); and

•           Ethical and practical feasibility of multiple trials.

FDA suggests that, in this pathway, the single trial will typically need to be “highly persuasive” and/or supported by “strong confirmatory evidence” given the absence of a second adequate and well‑controlled trial.

Examples of the confirmatory evidence that the Agency expects would be adequate include:

•           Prior adequate and well‑controlled trials of the same drug in closely related indications (e.g., different stages of a cancer; related infectious disease sites; diseases sharing a common precursor or pathophysiology).

•           Adequate and well‑controlled trials of other drugs in the same pharmacologic class, with closely similar mechanisms of action, comparable endpoints, and similar effects. Presumably, these could even be a competitor’s studies.

Alternatively, even if strong external or related‑trial confirmatory evidence is not available, FDA indicates that a single trial may still provide substantial evidence if that single study is a “highly persuasive” adequate and well‑controlled trial and there is early‑phase evidence (e.g., mechanistic, pharmacodynamic, or other early‑phase data) that can serve as the confirmatory evidence.

The guidance also notes that where clinical trial results are really that good (e.g., pronounced mortality reductions), conducting a second trial would be unethical or impractical.  That too may suffice as the confirmatory evidence.

B. More than one adequate and well‑controlled clinical investigation

The guidance also recognizes the traditional pathway of multiple adequate and well‑controlled trials as a robust means to demonstrate substantial evidence.  More than one clinical trial may be necessary to demonstrate substantial evidence of efficacy where:

•           A single trial is not representative of the intended treatment population;

•           The safety or benefit‑risk assessment requires additional exposure; or

•           Specific deficiencies in the single trial leave residual concerns.

C. Leveraging known effectiveness of an approved drug

Finally, the guidance also recognizes that, in some cases, sponsors can meet the substantial evidence standard without conducting any new adequate and well‑controlled effectiveness trials by extrapolating known effectiveness from an approved drug to:

•           A new population (e.g., adult to pediatric); or

•           A new dose, regimen, route of administration, or dosage form

provided such extrapolation is scientifically justified and legally permissible.

Regulatory Flexibility: When and How FDA May Tolerate Greater Uncertainty

The final section of the guidance discusses “regulatory flexibility.”  While the statutory effectiveness standard is constant and unchanging, how that standard is met may vary with clinical context.  Quite reasonably, drugs addressing particularly severe diseases, unmet needs, or rare conditions warrant more regulatory flexibility than others.

As with all draft guidances, this one is nonbinding and open for public comment.

Photo of Bexis

We’ve always been interested in litigation over efforts to access “drugs” that are not legally available due to lack of any FDA approval – whether that involves attempts to force drug companies to provide discontinued investigational drugs, “Right to Try” legislation, or as in today’s decision, attempts to broaden the “practice of medicine” to include not just off-label use, but no-label-at-all use.  The defendant physicians’ efforts failed in United States v. California Stem Cell Treatment Center, Inc., ___ F. Supp.3d ___, 2026 WL 2575467 (C.D. Cal. Aug. 28, 2026), and they were enjoined from engaging in “adipose tissue-derived stem cell therapies.”

The defendants – a couple of doctors with no relevant background (one was a urologist and the other a cosmetic surgeon, id. at *5) and their business entities – set up a system of “more than 100 clinics, located throughout the United States and other countries” that marketed what sound like biologics that had no FDA approval, clearance, or other form of marketing permission:  several  product[s] containing “stromal vascular fraction” (“SVF”), which is “manufactured from a patient’s adipose tissue, and then combined with other components.”  Id. at *4.  The defendants licensed their “intellectual property and equipment to manufacturer and administer” these products.”  Id. at *5.  They claimed that the products could treat “neurological, autoimmune, orthopedic, and degenerative medical conditions and/or diseases, including, but not limited to, cancer, arthritis, stroke, amyotrophic lateral sclerosis, multiple sclerosis, macular degeneration, Parkinson’s disease, and chronic obstructive pulmonary disease.”  Id. at *6 (abbreviations omitted).  These products were manufactured using a patient’s own “adipose tissue,” which was removed, manipulated in various ways, and returned to the patient’s bodies.  Id. at *8-9, 11-12.

Reading the opinion’s findings of facts sure makes it sound like the defendant’s operation was an elaborate scam, but that’s not what makes the decision blogworthy.  Most of the Blog’s readers hardly need to be told that, in the USA, completely unapproved drugs (or other FDA-regulated products) are, by definition, adulterated and misbranded.  The legally interesting (and thus blogworthy) parts of the lengthy California Stem Cell opinion are the decision’s analysis, and rejection, of the defendant’s various defenses.  We count nine of those. These defendants tried everything but ultimately got nowhere.

First, the defendants were not merely engaging in surgery and otherwise practicing medicine on their patients.  Instead, they were administering products that were both “drugs” and “biologics.”  2026 WL 2575467, at *18.  “Because Defendants’ ‘intended use’ of the . . . products is to treat or mitigate a variety of diseases and medical conditions” those “products are ‘drugs’ under the FDCA.”  Id. at *32.  They were also “‘biological products’ . . . because each is a “virus, therapeutic serum, toxin, antitoxin, vaccine, blood, blood component or derivative, allergenic product, protein, or analogous product applicable to the prevention, treatment, or cure of a disease or condition of human beings.”  Id. (citations and quotation marks omitted).

Second, the products had not undergone any “adequate and well-controlled studies,” so they were “new drugs.”  Id. at *19.  The defendants’ failure to comply with virtually any of the FDA’s current good manufacturing practices or to have any FDA-approved labeling meant that the products were both adulterated and misbranded.  Id. at *34-35.  “Failure to comply with even one CGMP regulation renders the drugs legally adulterated.”  Id. at *34.  It was “impossible to draft adequate directions for use because there is no scientifically valid evidence to show that the . . . products are safe or effective for any indication.  Id.

Third, that the defendants’ products were “autologous” did not somehow exempt them from FDA regulation:

Autologous biological products may be created for individual patients using the patients’ own cells.  Because these autologous biological products are subject to FDA regulation, CBER routinely reviews applications concerning such products.  Autologous biological products . . . can and should be produced “with consistent strength, quality, and purity. . . .  [T]he . . . products are intended for autologous use, which refers to the “implantation, transplantation, infusion, or transfer of human cells or tissue back into the individual from whom the cells or tissue were recovered.

2026 WL 2575467, at *20-21 (citations omitted).  The “products are intended for autologous use, which refers to the ‘implantation, transplantation, infusion, or transfer of human cells or tissue back into the individual from whom the cells or tissue were recovered.’”  Id. at *33 (quoting 21 C.F.R. §1271.3(a)).

Fourth, defendants could not get away with their claim that their completely unapproved products were being “studied” as medical procedures governed by state law.  Defendants concededly had never sought to conduct any FDA-regulated investigational drug or biologic clinical trial.  2026 WL 2575467, at *28 & n.21.  Far from regulating the defendants’ activities, the “California state health authorities worked together [with the FDA] to embargo and seize the [product] that Defendants were using to make the treatment” because it was “human experimentation.”  Id. at *30 (citations omitted).  Defendants could not simply “take something that is unapproved . . ., combine it with something that is approved, and use that to treat diseases and conditions” . . . because they are doctors and/or conducting clinical research.”  Id. at  *31 (citation omitted).

Defendants’ unauthorized experiment treating cancer patients with an unproven product containing live [smallpox] virus not only put the patients at risk, but also represented a risk to the public health because people in close contact with those patients could have been infected with live virus.

Id. at *35 (citation omitted).  Plaintiffs often purport to equate off-label use with medical “experimentation.”  California Stem Cell illustrates what true “experimentation” looks like.

Fifth, defendants could not avoid regulation by characterizing their activities as “patient-funded research.”  Id. at *36.  “A ‘sale in the strict sense’ does not have to occur to bring a physician within the reach of” the FDCA’s “held for sale” requirement.  Id. at *35 (citation omitted).  “[A] physician’s use of a device on a patient is covered by the statutory phrase ‘held for sale.’”  Id. (citation and footnote omitted).  That phrase “does not focus narrowly on whether a physician charges for the device (or drug) consumed in the course of treating the patient, but focuses more generally on the commercial nature of the transaction, actors, and products at issue.  Id. (citation and quotation marks omitted).

Defendants’ . . . products are “held for sale” within the meaning of [21 U.S.C. §]331(k).  Defendants are practicing physicians engaged in the business of providing [those] products to patients to treat a variety of diseases and conditions . . . .  Many patients pay thousands of dollars to receive a single . . . product, and some patients pay much more to receive multiple treatments . . . .  Defendants are commercial actors who market and offer their . . . products to patients for commercial purposes other than Defendants’ own personal consumption, thereby rendering the CSCTC products “held for sale” within the meaning of [the statute].

2026 WL 2575467, at *36.

Sixth, defendants’ claim that they were not in “interstate commerce” also failed.  “The final drug product . . . need not have been shipped in interstate commerce in completed form.”  Id. at *37.  It was sufficient that “only an ingredient is transported interstate.”  Id. at  (citation and quotation marks omitted).  Defendants could not escape by performing “procedures” only in one state.  “[E]ven purely local activities that are part of an economic class of activities that have a substantial effect on interstate commerce,” are legitimately subjected to the FDCA.  Id. (citation and quotation marks omitted).

[E]ven where the purported medical procedure occurred entirely within the state, the . . . product nonetheless had sufficient connection to interstate commerce to permit federal regulation under the Commerce Clause.  Both the constitutional and statutory requirements for interstate commerce are satisfied here.

Id. (citation and quotation marks omitted).

Seventh, the FDA’s “same surgical procedure exception” (21 C.F.R. §1271.15(b)) does not apply.  The “adipose tissue that Defendants remove from patients is not just modified, but actually dismantled and destroyed, via . . . various . . . processing steps.”  2026 WL 2575467, at *11.  The regulation’s “narrow exceptions” do not apply to the defendant’s wholesale processing of removed patient tissue.  Id. at *38.  Rather, tissue that is reimplanted “must be . . . in the form removed from the patient for the SSP exception to apply.”  Id. at *39.  Defendants did much more than the “minimal manipulation” allowed by the regulation.  Id. at *44-45.

Defendants’ establishments do not qualify for the SSP exception because Defendants remove from patients . . . adipose tissue − and following processing implant a [something] different . . . − a cellular product [that was] combined with other drug components. . . .  Defendants’ manufacturing process fundamentally alters the adipose tissue that they removed from the patient, and further alters the physical and biological characteristics of the cells originally contained in the tissue.

Id. at *39 (footnotes omitted).  Defendants’ interpretation of the regulation was rejected because it:

would allow an establishment to remove any tissue from any part of a patient, perform any number and type of manufacturing steps on that tissue in relation to any purported surgical procedure . . ., inject the end product into any part of the patient, and then invoke the SSP exception as long as the end product contained one or more cells that were present in the original tissue − no matter how wildly different the end product might be.

Id. at *40.  Nor was the FDA’s construction of §1271, a “surprise” or any change from the Agency’s longstanding view of how its regulation operated.  Id. at *43-44.  “Under its enforcement power, FDA has absolute discretion as to whether to bring an enforcement action” and “ha[d no obligation to refrain from enforcing against Defendants’ violations of the FDCA.”  Id. at *44 (citation omitted).

Eighth, defendants were “manufacturers” under the FDCA, despite their arguments that “they are simply physicians who are practicing medicine and performing surgery.”  Id. at *47.  This was not a case of mere off-label use.  The FDCA’s practice of medicine exemption, 21 U.S.C. §360(g), “does not authorize doctors to use unapproved drugs, even in their ‘practice of medicine.’”  California Stem Cell, 2026 WL 2575467, at *47 (citations omitted) (emphasis original).

[E]ven doctors must comply with FDCA requirements. . . .  Congress did not create a broad “practice of medicine” exception that allows physicians to do whatever they please.  While the FDCA was not intended to regulate the practice of medicine, it was obviously intended to control the availability of drugs for prescribing by physicians.

Id. (citations and quotation marks omitted).

The Blog has been supportive of physician off-label use since its inception.  What went on in California Stem Cell was not off-label use, but rather involved manufacturing practices that simply ignored any and all FDA good manufacturing practices:

Although Defendants claim they merely engage in “off-label uses” of drugs and other medical products, that argument fails because − as Defendants admit − the . . . products have not been approved by FDA for any use. . . .  Although off-label use allows physicians to prescribe lawful drugs for unapproved uses, off-label use of adulterated products is beyond the scope of the privilege  Thus, it is well-settled that while a physician may exercise professional judgment in the off-label use of unadulterated products, nothing in the FDCA or caselaw suggests that the use of adulterated products is ever permissible.

2026 WL 2575467, at *47 (citation and quotation marks omitted) (emphasis original).  Whatever the defendants’ may have subjectively believed they were doing, what they actually did was manufacture and sell unapproved drugs and biologics, making them subject to FDA regulation and prosecution.  Id. at *48-49.

Ninth, and finally, defendants could not invoke their “patients’ rights to control their own cells.”  Id. at *48.  No such “constitutional right” exists.  “[T]here is no constitutional right to receive unapproved drugs, regardless of their composition,” even for “terminally ill patients.”  Id. (citations omitted).  Nor did defendants’ activities fit within any “right to try” statutory provision. 

The[se] products do not meet the statutory definition of an “eligible investigational drug” that patients have a “right to try” because, inter alia, [they] have been placed on clinical hold, are not subject to an effective [investigational application], and have not been studied in a completed, FDA-approved Phase 1 clinical trial.  Thus, the FDCA’s “right to try” provisions decidedly do not permit Defendants to use their unapproved drugs to treat patients.

Id. (citations omitted).

California Stem Cell thus contains by far the most thorough treatment of issues related to the manufacture and use of entirely unapproved drugs that we have ever seen – far exceeding the prior record holder, Harborth v. State, ex rel. Dep’t of Workforce Services, 424 P.3d 1261 (Wyo. 2018), which we discussed here.  Other cases on the topic, such as United States v. Livdahl, 459 F. Supp.2d 1255, 1266-67 (S.D. Fla. 2005); Gaston v. Hunter, 588 P.2d 326, 351 (Ariz. App. 1978); and Retkwa v. Orenteich, 584 N.Y.S.2d 710, 713 (N.Y. Sup. 1992), are much older and contain even less analysis.

There’s even more.  Beyond the decision’s rejection of no fewer than nine (as we see it) various defenses to trafficking in unapproved drugs/biologics, California Stem Cell also contains a useful (and rather humorous) takedown of one of defendant’s “expert” witnesses as “unreliable and [who] should be disregarded.”  That witness – it is difficult to continue with the adjective expert, even in quotes – “admitted . . . that substantial portions of her expert report were not her own.”  2026 WL 2575467, at *16. 

When . . . questioned about the content of her expert report and opinions, she not only conceded that she had not authored substantial portions of her report, but also stated that she had no idea how those portions had come to be included in her report.  When confronted with a side-by-side comparison of her expert report in this case and [a different expert’s report in related litigation] − and questioned about their extensive similarities, which had been highlighted in yellow . . . – [she] testified that those portions had been merged with her report, that she had not written the additional portions, and that much of the text in the two reports was identical.

When asked whether she had, in fact, reviewed the scientific literature cited throughout the footnotes to her expert report upon which her opinions purportedly were based, [she] testified, “I had nothing to do with the writing of any of the things that you just highlighted or any of the references that you’re pointing to.  I had nothing to do with that.  I was surprised when I saw that merged together with what I had written. What I wrote was completely separate.”

Id. at *17 (citations omitted).  Wow!  That’s bad enough to qualify for one of our “stupid expert tricks” posts.

Photo of Michelle Yeary

Some courtroom lines never die. More than thirty years later, everybody still knows the glove, the struggle to pull it on, and the rhyme that went with it — “If it doesn’t fit, you must acquit.” Here’s the thing, though. Long before anyone was trying on gloves in Los Angeles, the Third Circuit’s Judge Becker had come up with “fit,” and the Supreme Court adopted it in Daubert. Scientific validity for one purpose is not necessarily scientific validity for another. A study of one population, one exposure, and one outcome doesn’t automatically prove anything about a different population, a different exposure, or a different outcome. If the science doesn’t fit the plaintiff, the expert doesn’t get to the jury. Or, updating the rhyme for Rule 702 — if the study doesn’t fit, you must exclude it.

Today’s case isn’t a drug or device case. It comes out of the Deepwater Horizon cleanup. But it’s a published Fifth Circuit decision affirming exclusion of plaintiff’s causation experts under Rule 702, and affirming summary judgment, by applying what the court called “widely accepted toxic tort causation principles.” DuFour v. BP Exploration & Production, Inc., ___ F.4th ___, 2026 WL 3215980, at *4 (5th Cir. Oct. 27, 2026). Toxic tort causation is our neighborhood, so we’re going to walk through it.

Plaintiff alleges that his exposure to toxic chemicals as a cleanup worker caused chronic worsening of his asthma. Id. at *1. Note the word “worsening.” Plaintiff had asthma long before the spill, going back to childhood. Id. at *7. That fact is important. To prove causation, plaintiff offered two experts, an epidemiologist and one of his treating pulmonologists. The district court denied his request for a Rule 702 evidentiary hearing, excluded both experts, and granted summary judgment. Id. The Fifth Circuit affirmed.

The parties agreed that general maritime law governed. That raised the question of what causation looks like in a maritime toxic exposure case. Plaintiff made it easy. He conceded that “[w]ithout an expert, [he] could not prove the central issue in this litigation – the level, extent, and duration of his exposure to harmful substances.” Id. at *2. He also agreed he had to prove “(1) the scientific knowledge of the harmful level of exposure to a particular chemical, and (2) that he was exposed to such quantities.” Id. That’s dose. That’s the ballgame in any toxic tort case. That, plus lots of precedent, made it easy for the court to find that toxic exposure injuries with multiple possible causes require expert testimony. Id. at *3. The court then held that the general-causation and specific-causation framework that has governed non-maritime toxic torts for decades is part of maritime law, either because enough decisions have established it or because the federal courts have created that rule. Id. at *4.

Plaintiff’s first argument was that the district court abused its discretion by not holding a live evidentiary hearing. His theory was that the case was scientifically complex and that his experts deserved a chance to explain away whatever deficiencies the court saw. Id. at *5. The Fifth Circuit disagreed. Trial courts have “considerable leeway” in deciding how to assess reliability. The district court had thorough briefing, depositions, declarations, and expert reports. It also had the benefit of its familiarity with the Deepwater Horizon litigation. That was enough. Id. at *6. A Rule 702 hearing isn’t a second chance to write the report you should have written the first time. Plaintiffs don’t get unlimited mulligans.

The court began its Rule 702 analysis with dose-response. It noted that even if Mississippi law applied instead of maritime law (remember that point when we get to the dissent), the Mississippi Supreme Court has affirmed exclusion of an expert whose studies did not provide a dose-response relationship that allowed the plaintiff’s exposure to be compared with the exposures of study subjects. Id. at *6 (discussing Watts v. Radiator Specialty Co., 990 So. 2d 143 (Miss. 2008)).

Expert No. 1: Plaintiff’s epidemiologist relied on three studies. They didn’t examine the same exposure-disease relationship. They didn’t address the same conditions. And none of them “appear[ed] to address the specific worsening of preexisting chronic asthma that is the central issue in [plaintiff’s] case.” Id. They weren’t even based on the same exposures. One focused on inhalation; the other two didn’t distinguish between exposure routes. Id.

The main study, referred to as the Lawrence Study, had two problems, either of which would have been enough. First, the expert himself acknowledged possible “misclassification and over reporting of asthma due to the utilization of self-reported wheezing in the case definition.” When self-reported wheezing was removed from the definition, the results were “largely attenuated.” Id. at *7. So, the association depended heavily on people saying they wheezed.

Second, and more importantly, the study excluded anyone who reported an asthma diagnosis before the spill. Plaintiff had asthma since childhood, so he would have been excluded. As the district court put it, and the Fifth Circuit agreed, that “demonstrates that he is not on common footing with any participant in that Study, and as such [the expert] did not reliably apply the Lawrence Study to this case.” Id. That’s fit in its purest form. The study was built for someone else’s hand.

Then there was the problem of symptoms versus disease. The dose-response the expert described concerned “the reporting of symptoms, not the diagnosis or worsening of asthma or any other chronic conditions.” Id. The expert conceded that wheezing and shortness of breath are symptoms of asthma but “they’re not specifically asthma worsening,” and that “not everybody who wheezes or has shortness of breath, et cetera, has asthma.” Id.

Next, statistical significance. Plaintiff (and the dissent) leaned hard on the study’s p-value. However, the fact that results are statistically significant “says nothing about the magnitude of any association . . . or about the biological, clinical, or public health importance of the finding.” Id. Significance “cannot make up for” the study’s other limitations. Id. at *8. The study did report a relative risk above 2.0 for some categories of workers, but the expert never tried to show that plaintiff would have fallen into any of those categories, even setting aside that the study excluded him. Id. A statistically significant result for a group you’re not in tells the jury nothing about you.

The bottom line — the expert offered no “reasonable basis for using respiratory symptoms alone as a proxy for chronic asthma.” Jumping from increased respiratory symptoms to the conclusion that the cleanup was a substantial factor in worsening plaintiff’s asthma “constitutes ipse dixit testimony and leaves too great of an analytical gap between the data and his causation opinion.” Id. Not an abuse of discretion. Too big a leap.

Expert No. 2: Plaintiff’s treating pulmonologist didn’t prepare an expert report. The district court ruled that a treater who doesn’t provide a report is limited to the facts and opinions in his disclosed medical records, meaning what he actually recorded while treating the patient. If those records don’t say plaintiff’s condition was caused by a particular substance, the treater can’t offer that causation opinion later in litigation. Id. at *8. Plaintiff didn’t rely on the records. He tried to use causation opinions the treater gave at his deposition, which the district court held “would plainly circumvent Rule 26’s written report requirement.” Id. (The dissent points to a chart note saying plaintiff “may have” work-related asthma. Id. at *10 (dissent). “May have” is something a doctor writes down to look into, not an opinion that something more likely than not caused the injury.) Whether treaters must file reports for opinions formed for litigation is not a settled issue in the Fifth Circuit, so the appellate court affirmed the exclusion on other grounds. Id. at *8 n.69.

Those other grounds were solid. Even if the opinions had been properly disclosed, nothing in the record showed the treater was qualified to opine on what the substances at issue do to the human body, in general or to plaintiff specifically. Id. at *9. He offered no explanation for ruling out other causes. He relied on timing alone. He admitted he didn’t know much about the substances involved and did no research into what might have worsened plaintiff’s asthma. Id. That’s ipse dixit again. With both experts out, summary judgment followed.

There was a dissent, so we’ll address it, briefly. The dissent’s central theory is that “[w]e cannot uphold the district court’s Rule 702 ruling without first identifying the relevant substantive law. That is because Federal Rule of Evidence 702 and Daubert depend on the substantive law applicable to the plaintiff’s claim.” Id. at *10. Then comes a hypothetical about a serial spider-bite victim, from which the dissent concludes that “[d]istrict courts must reach different Daubert rulings when applying State A’s law and State B’s law to factually identical spider-bite suits.” Id. at *12. And because the dissent says we don’t know which law governs here, it would not uphold the exclusion. Id.

We don’t buy it. Rule 702 shouldn’t mean different things in different federal courtrooms, whether the court is sitting in admiralty or in diversity. Here’s how the division of labor actually works. State substantive law decides what must be proved. Rule 702 decides admissibility. State law sets the elements, the causation standard, whether expert proof is required, any degree-of-certainty requirement, and whether the evidence is sufficient. Federal law decides whether an expert’s opinion is admissible in the first place. Which state’s law governs doesn’t change how Rule 702 is applied.

The text of the Federal Rules backs this up. When the drafters wanted state law to control an evidentiary question in civil cases, they said so expressly: Rule 302 (presumptions), Rule 501 (privilege), and Rule 601 (witness competency) all defer to state law where state law supplies the rule of decision. Rule 702 has no such clause. It’s a uniform federal standard that should be applied uniformly. 

Yes, “fit” asks whether the testimony will help determine “a fact in issue,” and the substantive law identifies those facts. But the fact in issue here was whether cleanup exposure worsened plaintiff’s preexisting asthma. That’s the same question under maritime law, Mississippi law, or any other law we can think of. No jurisdiction makes causation optional, and plaintiff conceded he needed expert proof of dose. Whether an expert reliably connected his data to that fact doesn’t depend on which body of tort law supplies the rule of decision. And as the majority pointed out, Mississippi applies Daubert-based dose-response principles anyway. Id. at *6. The choice-of-law question the dissent raises makes no difference to the result.

The dissent also complains that no court has ever held “an expert cannot rely on scientific studies unless the plaintiff himself qualified as a subject of the study.” Id. at *13 (dissent). That isn’t what the court held. The holding was that the expert never bridged the gap between the population the study examined and the plaintiff in the case, and never even tried to place plaintiff in the categories that showed elevated risk. Id. at *8. That’s a Rule 702(d) reliable-application failure, plain and simple. Finally, the dissent called the new Reference Manual “biased and scandal-ridden.” Id. The majority’s answer was low-key.  The quoted passage on statistical significance is nearly identical to language in the “uncontroversial third edition,” and nobody has criticized either edition’s treatment of statistics. Id. at *7 n.67. Whatever anyone thinks of the fourth edition, the principle that statistical significance isn’t the same thing as relevance is basic statistics.

Just to recap the highlights of DuFour applicable to drug and device cases: (i) a Rule 702 hearing isn’t required when the record is fully developed; (ii) an expert’s studies must address the same exposure, the same outcome, and a population the plaintiff actually belongs to; (iii) symptoms are not a stand-in for disease without a reliable scientific bridge; (iv) statistical significance can’t rescue a study that answers the wrong question; and (v) a treater who skips the report is limited to what’s in the chart, and a timeline is not a methodology. All of that applies whenever a plaintiff’s expert offers a study of somebody else’s disease in somebody else’s population.

Back to the glove. That jury was told that if it didn’t fit, they had to acquit. Here, the district court tried plaintiff’s science on and found it was made for a different hand. A different population, a different outcome, a different question. The Fifth Circuit agreed. The study didn’t fit, so it was excluded.

Photo of Bexis

We held off blogging about Navarro v. Walgreens Boots Alliance, Inc., 2026 WL 2355975 (Mag. E.D. Cal. Aug. 13, 2026) (“Navarro I”), in the hope that this absurd magistrate decision would be reversed by the district court.  That didn’t happen – see Navarro v. Walgreens Boots Alliance, Inc., 2026 WL 2924631 (E.D. Cal. Sept. 29, 2026) (“Navarro II”).  So, let’s dive into this ridiculous and atextual result.

This litigation is another example of the out-of-control OTC drug consumer class action environment enabled by an obsolete “zombie” presumption against preemption decision from the Ninth Circuit, Astiana v. Hain Celestial Group, Inc., 783 F.3d 753 (9th Cir. 2015), abrogated on other grounds, Hartford Casualty Insurance Co. v. J.R. Marketing, LLC, 353 P.3d 319 (Cal. 2015). We discussed this aspect of Astiana in our general OTC preemption post, here.

The Navarro I magistrate decision expressly relied on Astiana in holding that the plaintiff could ignore the express preemption clause directly governing OTC drugs, 21 U.S.C. §379r(a), in reliance on general FDCA “misbranding” language, so that plaintiffs could pursue warning claims directly contrary to what the FDA’s preemptive monograph allowed:

Chapter V of the FDCA generally prohibits the misbranding of drugs. 21 U.S.C. §352.  A drug is misbranded if its labeling is false or misleading in any particular. Id.  The Ninth Circuit has made clear that state causes of action alleging that a label is “false or misleading in any particular” are not generally preempted by the FDCA even if there is not a specific FDA regulation prohibiting the statement.  See Astiana v. Hain Celestial Grp., Inc., 783 F.3d 753, 758-59 (9th Cir. 2015) (in context of cosmetics). . . .  That is, OTC acne medications, among other requirements, must not contain labeling that is false or misleading in any particular.

Navarro I, 2026 WL 2355975, at *22.

But as we discussed in our prior post, the 2015 Astiana decision explicitly relied on a presumption against preemption in narrowly construing identical express preemption language in §379s (concerning cosmetics), to allow generalized “misbranding” claims to survive preemption:

In analyzing express preemption, we start with the assumption that the historic police powers of the States were not to be superseded by the Federal Act unless that was the clear and manifest purpose of Congress.  The FDCA proscribes any . . . labeling that is false or misleading in any particular.

783 F.3d at 757.

That presumption no longer exists in cases such as Navarro that involve express preemption.  One year after Astiana, the Supreme Court outright abolished any purported presumption against preemption in express preemption cases:

The plain text of the [preemption clause] begins and ends our analysis. . . .  And because the statute contains an express pre-emption clause, we do not invoke any presumption against pre-emption but instead focus on the plain wording of the clause, which necessarily contains the best evidence of Congress’ pre-emptive intent.

Puerto Rico v. Franklin California Tax-Free Trust, 579 U.S. 115, 125 (2016) (citations omitted).

Since PR v. Franklin, the Ninth Circuit has repeatedly rejected attempts to rely on any presumption against preemption in express preemption cases.  In Hollins v. Walmart, Inc., 67 F.4th 1011 (9th Cir. 2023), an FDCA case, the court held – directly contrary to Astiana:

[A] state-law misbranding claim that would allow a state to impose requirements different from those permitted under the FDCA is preempted.  “[W]e do not invoke any presumption against pre-emption” where, as here, “the statute contains an express pre-emption clause.”

Id. at 1016 (quoting PR v. Franklin; other citations and quotation marks omitted).  Numerous other Ninth Circuit decisions likewise recognize the demise of any presumption against express preemption.  See KalshiEX, LLC v. Assad, 190 F.4th 1098, 1110 (9th Cir. 2026) (“we decline to apply any presumption against preemption and instead focus on the plain wording of [the statutory preemption clause]”) (citation and quotation marks omitted); American Apparel & Footwear Ass’n, Inc. v. Baden, 107 F.4th 934, 939 (9th Cir. 2024) (“the plain wording of the express preemption clause ‘necessarily contains the best evidence of Congress’[s] preemptive intent’”) (quoting PR v. Franklin); Assurance Wireless USA, L.P. v. Reynolds, 100 F.4th 1024, 1032 & n.4 (9th Cir. 2024) (“When a statute contains an express preemption clause, we look only to the text ‘without any presumptive thumb on the scale for or against preemption.’”; “[t]he presumption against preemption does not apply when there is an express preemption clause”) (quoting California Restaurant); California Restaurant Assn. v. City of Berkeley, 89 F.4th 1101, 1050 (9th Cir. 2024) (“we apply [statutory] textual analysis without any presumptive thumb on the scale”) (citation and quotation marks omitted); National Railroad Passenger Corp. v. Su, 41 F.4th 1147, 1153 n.1 (9th Cir. 2022) (“because the statute contains an express pre-emption clause, we do not invoke any presumption against pre-emption”) (quoting PR v. Franklin); Nacarino v. Kashi Co., 77 F.4th 1201, 1204 (9th Cir. 2023) (“[w]hen, as here, ‘the statute contains an express pre-emption clause, we do not invoke any presumption against pre-emption but instead focus on the plain wording of the clause”) (quoting PR v. Franklin); R.J. Reynolds Tobacco Co. v. County of Los Angeles, 29 F.4th 542, 553 n.6 (9th Cir. 2022) (quoted in California Restaurant); Webb v. Trader Joe’s Co., 999 F.3d 1196, 1202 (9th Cir. 2021) (“[w]here the intent of a statutory provision that speaks expressly to the question of preemption is at issue, we do not invoke any presumption against pre-emption”) (citation and quotation marks omitted); Connell v. Lima Corp., 988 F.3d 1089, 1097 (9th Cir. 2021) (quoting PR v. Franklin); International Brotherhood of Teamsters, Local 2785 v. Federal Motor Carrier Safety Administration, 986 F.3d 841, 853 (9th Cir. 2021) (same as Nacarino); Atay v. County of Maui, 842 F.3d 688, 699 (9th Cir. 2016) (quoted in Webb).

Thus, the Ninth Circuit, following controlling Supreme Court precedent, has massively repudiated the presumption against preemption reasoning that led to the Astiana court’s creation of a “misbranding” exception to §379r(a)’s express language that all state tort claims that are “different from or in addition to, or that [are] otherwise not identical with” the FDA’s OTC drug requirements are preempted and thus of no effect. Nonetheless, Navarro I went down the Astiana road and reached an absurd result.

The Navarro class action plaintiffs brought state-law claims for unfair competition, false advertising, consumer protection violations, breach of express and implied warranties, and unjust enrichment.  Nobody was hurt; they sought recovery only of purported economic harms.  Navarro I, 2026 WL 2355975, at *4-5.  The liability allegations are familiar – the defendant allegedly sold an OTC products (acne treatment) that purportedly “contain and/or degrade into dangerously high levels of benzene” which plaintiffs claim, in this context, is an “impurity, and illegal adulterant.”  Id. at *1.

The trouble is, as many other courts – particularly outside the Ninth Circuit − have found, is that the FDA issued a “drug monograph” for these products that allowed use of the ingredient in question (benzol peroxide (“BPO”)) as an active ingredient in these medications and did not require any warnings about that ingredient decomposing into benzene.  Id. at *8 (“The Acne Monograph expressly permits BPO to be an active ingredient in these products in an amount from 2.5 to 10 percent.”).  Plaintiffs claim that this can happen, if one literally cooks the product at “122 degrees Fahrenheit for 18 days” as the plaintiff-friendly Valisure litigation lab did.  Id. at *3.  That’s a hotter stretch than Death Valley in the summertime.

To avoid preemptive effect of §379r(a)’s preemptive prohibition of state labeling requirement “different from or in addition to, or that [are] otherwise not identical with”, the magistrate in Navarro I jerry rigged a legal contraption that began with a sentence in the monograph that this type of product “is not misbranded if it meets each of the conditions in this subpart and each general condition established in §330.1 of this chapter.“  2026 WL 2355975, at *9 (emphasis original).  But §330.1 provides only that a nonconforming product is “liable to regulatory action.”  Id.  Nothing in that section says anything about preemption or state law.  The magistrate then concluded that, because the monograph “in turn incorporates Chapter V labeling requirements,” id. at *10, it can use the adulteration/misbranding dodge to avoid the express terms of §379r(a).  “[A] theory of misbranding, pursuant to state-law, may sufficiently parallel the FDCA’s requirements under Chapter V.”  Id.

Navarro I also had avoid the Supreme Court’s recent Durnell reaffirmation of the preemptive effect of similar preemption language.  It did so with a bootstrap argument relying on the same rationale it had just cobbled together: 

The Court has reviewed the case and finds that it is not apposite to the case at bar.  Critically, Durnell did not involve the interplay between a monograph and preemption statute.  Rather, in Durnell − a case involving pesticides under the Federal Insecticide, Fungicide, and Rodenticide Act − the issue before the Court involved only a preemption statute.

Id. at 10 (citation omitted).  That is an excuse rather than a reason.

As the Supreme Court recognized, the language of FIFRA and the FDCA is quite similar.  Durnell held that “different from” and “in addition to” preemption language (part of §379r(a) as well) mean just that

In sum, federal law requires [defendant] to sell [its product] with the label that EPA approved . . . − that is, the label without a cancer warning.  [Plaintiff]’s state tort claim, by contrast, would require [defendant] to add a cancer warning to its labels.  That [state]-law requirement is “in addition to” and “different from” [defendant]’s federal-law labeling obligations.

Monsanto Co. v. Durnell, 609 U.S. 371, 385 (2026).  No “interplay” is necessary under Durnell. The language state law purportedly demanded was “different from” and “in addition to” what the responsible regulatory agency required.  Therefore, it was preempted.  Nor could “general” misbranding allegations overcome the “specific requirements” applicable to the particular product at issue:

[T]hat argument operates at far too high a level of generality and disregards the central and comprehensive role that [agency] performs in making labeling determinations under [the statute’s] registration provisions.  Looking at only {the statute’s] general standard for misbranding rather than the specific requirements imposed under federal law would nullify [the statute’s] preemption clause and the uniformity that Congress sought for safety warnings on pesticide labels.

Id. at 388.  Durnell thus does not admit of Navarro I-style sophistry.  “If FDA’s premarket approval . . . preempted additional state-law requirements, so too must EPA’s registration.”  Id. at 387.

To allow plaintiffs’ adulteration claim, Navarro I engaged in similarly makeshift analysis that invoked a general FDCA provision (21 U.S.C. §351(b)) and something from out in left field – the United States Pharmacopœia.  Rather than make this post even longer by explaining this ourselves, we instead offer the analysis in a more recent decision in the same litigation that thoroughly debunks this contrived argument:

The pathway from the FDCA to the USP standards is found at 21 U.S.C §351(b), which states that a drug is adulterated “[i]f it purports to be or is represented as a drug the name of which is recognized in an official compendium, and its strength differs from, or its quality or purity falls below, the standard set forth in such compendium.” Plaintiffs then claim that “[u]nder USP General Chapter 467, benzene is limited to 2 ppm in pharmaceuticals.”

The problem is that the USP General Chapter 467 limit applies to benzene when used as a “Residual Solvent.” Chapter 467 defines residual solvents in pharmaceuticals “as organic volatile chemicals that are used or produced in the manufacture of drug substances . . ., or in the preparation of drug products” and that “are not completely removed by practical manufacturing techniques.”. . . .

Plaintiffs’ allegation . . . is that BPO, through a multi-step process, “decompose[s]” into benzene. . . .   In Plaintiffs’ words, benzene is not directly “produced” in the manufacturing process and, therefore, does not fall under USP Chapter 467’s definition of a residual solvent.

Lastly, the Court makes a final observation on the repeated use of the word “should” throughout USP Chapter 467, which causes its statements to be read as recommendations not binding requirements. . . . Even if benzene somehow met USP Chapter 467’s definition of a “residual solvent,” the Court would find that USP Chapter 467 . . . is non-binding guidance.

In re L’Oreal USA Benzoyl Peroxide Products Litigation, ___ F. Supp.3d ___, 2026 WL 2719653, at *10 (S.D.N.Y. Sept. 15, 2026) (citations omitted) (emphasis added).  Plaintiffs’ allegations in this litigation about where benzene supposedly comes from are identical.  E.g., Navarro I, 2026 WL 2355975, at *2 (“Plaintiffs allege that the BPO in products made and sold by [defendant] degrades over time to directly form benzene.”).  The USP aspect of the opinion is another slipshod excuse to avoid preemption.

The result is facially absurd.  The controlling FDA monograph has consistently allowed the active ingredient BPO as an active ingredient in the subject product.  The same monograph does not mandate any labeling concerning “benzene” in any form.  Congress’ express preemption language flatly precludes any purportedly state-law claim that is “different from or in addition to, or that is otherwise not identical with” the FDA’s requirements.  Nonetheless, by relying on general “adulteration”/”misbranding” provisions, Navarro I allowed exactly that – state law claims demanding “benzene” labeling – in short “parallel” claims that are in no way parallel to what the legally binding FDA monograph requires.

We were inclined to dismiss Navarro I as the misguided work of a pro-plaintiff magistrate, but we were wrong.  In a three-page order, Navarro II, 2026 WL 2924631, adopted Navarro I in full.  It refused to follow either Durnell, Yousefzadeh v. Johnson & Johnson Consumer Inc., 184 F.4th 130 (2d Cir. July 30, 2026), or L’Oreal USA Benzoyl Peroxide, 2026 WL 2719653.  As we recently discussed, Yousefzadeh, specifically rejected the presumption-based preemption analysis in Astiana that allowed “misbranding” claims contrary to the FDA’s product-specific requirements.  In a single sentence, Navarro II declared Durnell “unavailing” without coming to grips with the Supreme Court’s specific rejection of misbranding/adulteration claims that negate product-specific agency requirements.  2026 WL 2924631, at *2.  It dispensed with Yousefzadeh in another single sentence, based on the unelaborated “facts and circumstances of this case.”  Id.  Finally, it flatly rejected the holdings of the MDL court in L’Oreal USA Benzoyl Peroxide:

Finally, defendant’s post-objection supplemental authority, In re L’Oréal USA Benzoyl Peroxide Products Litigation, 2026 WL 2719653 . . ., is non-binding and in any event, is not a basis to depart from the instant findings and recommendations.  L’Oréal’s determination that the FDA considered and resolved BPO related benzene risk through the monograph and GRASE findings is belied by the FDA’s subsequent benzene alerts and warnings, the independent [Valisure Death Valley Days] testing. . . .  Also, the L’Oréal Court’s rejection of the theory that benzene is a residual solvent for purposes of USP Chapter (467) did not consider BPO degradation as a potential solvent source, a matter discussed in detail by the magistrate judge.

2026 WL 2924631, at *2.  As our block quote from L’Oréal two paragraphs previous demonstrates, Navarro II’s last critique about failure to “consider” product “degradation” in connection with the plaintiffs’ “USP” argument is simply wrong.  L’Oréal, considered that claim in depth and found that degradation allegations inherently precluded reliance on the USP (which lacked force of law in any event).

We’ve already discussed in our prior post about Yousefzadeh that there is now an express circuit split between the Second and Ninth Circuits on the fundamental issue in Navarro, which is the use of generalized misbranding/adulteration allegations in OTC preemption cases to assert claims at odds with FDA product-specific requirements.  The Ninth Circuit’s Astiana-based approach is undercut by the Supreme Court’s subsequent abolition of presumptions against express preemption, while the Second Circuit’s view is bolstered by Durnell’s recent rejection of such claims in the FIFRA context.  Now it looks like the same circuit split has also arisen specifically in the BPO litigation – with Navarro rejecting the MDL decision in L’Oréal.

If ever a decision warranted interlocutory appeal to resolve controlling and fundamental judicial conflicts, the Navarro litigation is it.

Photo of Steven Boranian

A postal worker picks up a leaking bottle of drain cleaner stuck on a conveyor belt and suffers severe chemical burns.  The product was sold through an online marketplace and was made by a manufacturer that sells only bulk cases to hardware distributors.  This wayward bottle was shipped in the US Mail illegally by someone, but we are not sure exactly who. 

Who, if anyone, owed our unfortunate postal worker a duty of care?  According to Rimel v. Alabama Janitorial & Paper Supply, LLC, No. 23-cv-2020, 2026 WL 2593320 (W.D. Pa. Sept. 2, 2026), whoever shipped the bottle to the plaintiff surely owed her a duty of care, and that may or may not have been Alabama Janitorial. 

The online marketplace, however, owed no duty at all, as it was a mere e-commerce platform acting as an intermediary for a third-party vendor.

You might be thinking, great, but why do we care?  We care because pharmaceutical products are purchased and shipped though e-commerce platforms in increasing numbers, and we know that all pharmaceutical products have both benefits and risks.  It is not too far outside the realm of possibility that someone experiencing an alleged pharmaceutical side effect could look to an online marketplace for compensation.

The Rimel order is as broad a statement on this issue as we have seen, and the district court there convincingly ruled that an online marketplace owes no duty of care to purchasers of products from third-party vendors.  The evidence showed that the e-commerce website notified the seller of the purchase, processed payment, and updated the buyer on order status.  That’s all.  It never possessed, warehoused, or shipped the bottle.  The district court observed that Pennsylvania’s highest court has not addressed marketplace liability.  The Third Circuit certified the question to the Pennsylvania Supreme Court in Oberdorf v. Amazon.com Inc., 818 F. App’x 138 (3d Cir. 2020), but the case settled before the Supreme Court ruled.

Filling that gap, the district court relied on a “growing consensus” of federal decisions.  First among them was Fike v. Global Pharma Healthcare Private, Ltd., 741 F. Supp. 3d 265 (E.D. Pa. 2024), which we discussed here.  That case applied Pennsylvania’s five-factor duty test to find that online “middlemen” owed no duty.  Other cited cases ruled the same way—invoking the rationale that online marketplaces did not make, sell, or distribute defective products.  They also owed no duty to warn under Section 388 of the Restatement (Second) of Torts since that provision applies to product suppliers, which an online marketplace is not.  In the final analysis, none of these courts was willing to create a duty to control the actions of others or warn of dangers created by third parties. 

The district court in Rimel thus announced a broad rule:  “As a general matter, the Court holds that e-commerce platforms . . . do not owe any duty of care relating to products sold by third-party vendors on their websites.”  It grounded that holding in the Pennsylvania principle that “there is no duty to control the conduct of a third party to protect another from harm,” absent a recognized special relationship, citing Ritz v. Ramsay, 305 A.3d 1056, 1061 (Pa. Super. 2023). Finding no such relationship, the district court granted summary judgment for the defendant e-commerce website.

The district court granted summary judgment for the product manufacturer too, since the plaintiff did not claim any negligence in connection with the product itself—only its shipping.  It was undisputed that the product manufacturer did not ship the offending bottle though the US Mail.  That act occurred downstream, leaving no evidence in the record to hold the product manufacturer in the case. 

But who shipped the bottle?  We may never know, but Alabama Janitorial’s owner submitted a declaration stating that they did not ship this product to this plaintiff, so they should get summary judgment, too.  That declaration did not carry the day because there was evidence going the other way.  Because “the veracity of representations . . . is a quintessential jury issue,” this defendant’s motion was denied. 

The opinion offers a categorical no-duty holding under Pennsylvania law for online platforms that neither possess nor ship third-party goods.  It is not an opinion from the Pennsylvania Supreme Court, but it flows with a building weight of authority, and it strengthens a marketplace’s defense in Pennsylvania federal courts for sure.

Photo of Eric Alexander

This is actually the second time we have written on a decision from a case involving banana plantations.  The first involved violence against Colombian banana plantation workers by the paramilitary/terrorist group known by the initials of its Spanish name, AUC.  This one involves claims of various health effects from the application of a particular pesticide on various plantations in Central and South America.  Heavy stuff, so we will deflect by mentioning the beloved Caribbean ditty “Day-O,” also known as “The Banana Boat Song.”  If it were to start playing, you would probably be able to sing along with the familiar lyrics.  In one, it is said that “a beautiful bunch of ripe bananas . . . hides the deadly black tarantula.”  In fact, the only species of tarantula endemic to any Caribbean islands is neither black nor deadly to humans.  One of the recurring lyrics recounts stacking bunches of bananas so that the “tally man” can “tally [the] bananas.”  This references how the plantation worker would get the appropriate credit for the bananas he or she gathered.  Fun song that we have known since we were kids, but so what, you mumble.  Well, Patrickson v. Dow Chem. Co., No. SCWC-22-0000687, 2026 Haw. LEXIS 266 (Haw. Sept. 21, 2026), is about the standards for evaluation causation in toxic tort cases and the court made some elementary mistakes.

The Patrickson litigation has been going on for a long time.  The Blog previously included a 2015 Hawai’i Supreme Court decision in a post on cross-jurisdictional class action tolling.  That came more than a decade after the U.S. Supreme Court issued a 2003 decision in the litigation on the application of the Foreign Sovereign Immunities Act.  For a case this old, it is surprising that the standards for causation had not come up before.  How a Hawai’i state court would have jurisdiction over these claims and how they could possibly be timely—the plaintiff at issue on appeal sustained alleged injuries between 1971 and 1983 and literature was out by 1980 describing some association between the pesticide and reproductive harms—are perhaps even bigger mysteries.  We are also not sure the plaintiff (or his estate, because he died after filing) is the right one to sue for the alleged injuries, that they are compensable, or that the claims should have survived his death.  Setting all that aside, the pertinent facts are that the decedent worked for a subsidiary of one of the defendants in Costa Rica between 1971 and 1973, where he was allegedly exposed to a pesticide made by another defendant.  His job did not involve applying or distributing the pesticide and he never saw it applied, but he might have been around “sacks” of it in a warehouse if he sought shelter there during heavy rain.  His claimed injuries were “infertility” and other reproductive impacts based on 1) his wife having only four new pregnancies in the years following the relatively short employment at issue and 2) three of those new pregnancies ending in a miscarriage or spontaneous abortion.  (Plaintiff’s wife also had a child who was born and died of “an illness” in the year he started work, and they also adopted another child.)  After the case eventually proceeded to discovery, defendants won summary judgment on the lack of proof of exposure and excluded plaintiff’s causation expert, largely because he assumed there had been sufficient exposure to cause the claimed reproductive injuries.

The intermediate appellate court reversed, and an appeal to the Supreme Court of Hawai’i followed.  The Patrickson decision quickly went off the rails.  It took two separate dives into the court’s prior endorsement of the alternate market share theory of liability on a certified question from the Ninth Circuit in Smith v. Cutter Biological, Inc., 823 P.2d 717 (Haw. 1991).  2026 Haw. LEXIS 266, *2-4 & 16-23.  Smith may be noteworthy because it is the only high court decision we know that applied this silly theory in the context of a prescription medical product other than DES, but even Patrickson said up front that the theory was “inapposite” to the plaintiff’s claims because the identity of the manufacturer of the pesticide that plaintiff’s employer used was known.  Id. at *3.  Market share theories have nothing to do with single manufacturer cases or cases where plaintiffs have proof of exposure to a specific defendant’s product, which plaintiff claimed to have in Patrickson.  They are for cases where plaintiffs cannot satisfy basic product identification requirements applicable to all product liability cases.  So, what was the point of talking about Smith so much?  Patrickson’s circuitous reasoning was that Smith showed Hawai’i could make up new causation standards to help “innocent plaintiffs” recover against someone with money.  Id. at *3, 18 & 21.  But the plaintiff in Patrickson had identified defendants and did not need some new causation theory any more than the typical plaintiff who sues a prescription drug company over an injury alleged sustained from taking its drug.  Hawai’i already has causation standard for such a product liability case.  It also has its own version of Daubert and a Rule 702 that is patterned after an older version of Federal Rule of Evidence 702.  Id. at *28-30.  Using a prescription medical product case as the core reason why the court had to devise a “causation standard in toxic tort cases” or “how causation should be addressed in mass toxic tort cases” seems like poor justification.  Id. at *12 & 22.  If plaintiff cannot make out product liability claims because of lack of proof, then the solution should not be to relax those standards for an ill-defined subset of product liability cases.

Still, some of the discussion of what a plaintiff has to prove was really basic.  After looking at various cases and Restatement provisions, Patrickson declared that each toxic tort plaintiff has to prove exposure, general causation, and specific causation.  Id. at *23-24.  Everyone knows those are requirements in all but the simplest product liability cases (e.g., the injury is claimed from a blunt force trauma).  The only places where that is not clear in cases involving the ingestion of drugs, the implantation of a medical device, or exposure to an allegedly harmful chemical are where courts get confused about there being two kinds of causation.  Good for Patrickson to recognize the obvious.  Also, Patrickson noted that “expert testimony should be required to establish causation in toxic tort cases, just as in medical negligence cases, given the special difficulty these cases present in proving a connection between a toxic substance and the development of a specific disease with a potentially significant latency period.”  Id. at *25.  Again, this is the law everywhere for drug, device, and chemical exposure cases, but it is good to reinforce.  (Also, plaintiff did not claim any “significant latency” period in the alleged impact on his reproduction, so we do not know why latency was raised.)  Then things got, well, bananas as Patrickson essentially undercut the basic principles it had reaffirmed.

The court stated that it “declined” to require a plaintiff “to provide expert testimony quantifying the exposure level or dosage to meet their burden of production.”  Id. at *26 (emphasis in original)  Instead, after a discussion that strayed into unrelated issues, the court held “that a plaintiff should offer some evidence ‘that he or she was exposed to the toxin at issue, including the estimated amount and duration of exposure.’”  Id. at *28 (quoting Lowery v. Enbridge Energy Ltd. P’ship, 898 N.W.2d 906, 914-15 (Mich. 2017) (concurrence, Markman, C.J.)). First, those are not actually different standards.  Second, what the concurrence in Lowery actually described was the importance of the quantification of exposure in the causation analysis:  “Accordingly, a plaintiff’s evidence of general causation should be tailored to the estimated amount and duration of exposure at issue to enable the fact-finder to reasonably conclude that exposure to the defendant’s toxin in the amount and duration alleged is capable of causing the alleged injury.”  898 N.W.2d at 914 (emphasis in original).  In that case, the concurrence wrote, the “plaintiff’s general-causation evidence falls short because it fails to show what level and duration of exposure to VOCs can cause the severity and duration of coughing and vomiting that is necessary to cause a gastric artery to avulse.”  Id. at 919 (emphasis in original).  In other words, if the general causation evidence is going to tie to the specific causation evidence, then the quantification of the plaintiff’s exposure must be relevant to what the studies show.  When it comes to expert evidence, there is a word for that concept:  fit.  The plaintiff’s experts’ opinions on causation must fit the case, or “reflect[] a reliable application of the principles and methods to the facts of the case” in the language of current Fed. R. Evid. 702(d).

Several of the authorities cited in Patrickson as the court tried to justify lowering the bar on causation make it clear that fit on the level of exposure or dose is important.  For instance, In re Hanford Nuclear Rsrv. Litig., 292 F.3d 1124 (9th Cir. 2002), which Patrickson cited as supporting the proposition that both general and specific causation could be established without quantifying the level of exposure, actually concluded with, “On remand, the district court should assess the plaintiffs’ proffered expert testimony as it relates to the generic causation inquiry, i.e., whether the radiation released from Hanford has the capacity to cause the illnesses alleged by plaintiffs.”  That requires quantification and fit.  If you go to the Federal Judicial Manual on Scientific Evidence, which is widely cited in the cases that Patrickson cited, includes extensive discussions about how the plaintiff’s general and specific causation evidence has to match on the dose or exposure.  Just going off of subject headings, within the section on “Specific Causal Association Between an Individual’s Exposure and Onset of Disease” are subsections on “Was the Plaintiff Exposed to the Substance, and If So, Did the Exposure Occur in a Manner That Can Result in Absorption into the Body?” and “When Exposure to the Substance Is Associated with the Disease, Is There a Benchmark Dose or No Observed Adverse Effect Level, and If So, Was the Individual Exposed Above This Level?”  Pretty obvious.  We could go on.  We could also say more about how the issues of whether epidemiologic studies are required to prove general causation, whether they must show a greater than 2.0 relative risk, and what differential diagnosis has to do with causation do not have much if anything to do with what plaintiff has to who about the “estimated amount and duration of exposure” to the substance that allegedly harmed her.  2026 Haw. LEXIS 266, *26-28.  The result was a fuzzy standard disconnected from the authorities discussed and basic principles of science.

The application to the evidence in Patrickson that the circuit court had considered in granting summary judgment made things worse. 

The circuit court was requiring a certain level of exposure when it [excluded the causation expert and granted summary judgment]. But we conclude that DiBartolomeis’s expert testimony met the touchstone of relevance and reliability, and that the ICA correctly held that the circuit court abused its discretion by excluding that evidence.

Id. at *29.  The basis for the expert’s assumption of exposure was the plaintiff’s deposition testimony about mere possibilities of some uncertain level of exposure and a study that raised a mere possibility of a relationship to spontaneous abortions in agricultural workers with documented exposure.  According to Patrickson, these “do provide circumstantial proof to facilitate a reasonable inference as to exposure and specific causation.”  Id. at *31.  It further noted that it could be sufficient for an expert’s opinion about causation to say “that a toxic substance ‘might’ be or ‘probably’ was the cause of the injury.”  Id. at *33.  There are a bunch of things wrong with these two statements.  Somewhat ironically, the court quotes the same passage from the referenced study twice, but does so inconsistently:  “[I]t is plausible that DBCP is causing an increased rate of spontaneous abortion in the exposed population through its effect on spermatogenesis” and “[I]t is possible that DBCP is causing an increased rate of spontaneous abortion in the exposed population through its effect on spermatogenesis.”  Id. at *8 & 27 n.7.  Just as “plausible” and “possible” mean different things, opinions on merely possible (“might be”) causation and likely (“probably was”) causation are different.  The former is inadmissible and the latter might be admissible if it is otherwise reliable and relevant.  In addition, a study on agricultural workers in Israel cannot be proof of past exposure for a plaintiff who once was an agricultural worker in Costa Rica.  The study could be part of a reliable opinion on general causation, but it would not support specific causation unless the “the estimated amount and duration of exposure” shown in the plaintiff—which Patrickson said it required—was relevant to what the study showed increased the risk of the same alleged injury.  Allowing expert testimony on disconnected possibilities does not fulfill the court’s role as a gatekeeper, which Patrickson also emphasized:  “[T]rial courts should exercise their gatekeeping function to ascertain whether proposed expert testimony meets the touchstones of relevance and reliability.”  Id. at *28-29.  With all these inconsistencies, it is hard to say that the Patrickson court articulated a standard at all.

This brings us back to our lyrical detour.  You need proof of exposure and general causation evidence that matches up with the specific causation issues in the case.  If the allegation is that the plaintiff died from a black tarantula bite while on a Caribbean island, then there would need to be proof of a bite and there would need to be expert evidence that the bite was capable of producing death.  There are no black tarantulas native there and the pink ones do not have venom that can kill a human, apparently,  Exposure needs to be quantified so you can see how it fits with the general causation evidence.  You need to tally up your bananas in whatever measure is used for banana counting for purposes of payment.  Or you need to count your tarantula venom exposure by whatever measure of tarantula venom (or its neurotoxin components) is used in the studies.  There are surely less circuitous analogies to what should be included in a causation analysis in a case where an injury is claimed from exposure to a substance.  The disjointed analysis in Patrickson, however, just makes us want to go home.

Photo of Stephen McConnell

We have covered several vaccine cases over the last few months, but Berenson v. Gottlieb et al., 2026 WL 3135334 (2d Cir. Oct. 20, 2026), is of a different sort. The plaintiff is a fairly famous writer who once worked for The New York Times, has written several books, and is now characterized by the court as an “independent journalist.” In his appellate brief, the plaintiff characterized himself as “a vocal critic of the public policy response to COVID-19, questioning the underlying evidentiary basis for business and school closures.” 

Today’s case arose when the plaintiff sued several government officials, the government, and the CEO of a vaccine manufacturer, claiming that they conspired to suppress his views expressed on social media. The specific claims were for (1) violation of the First Amendment; (2) a conspiracy to suppress the plaintiff’s speech in violation of 42 U.S.C. Section 1985(3); and (3) tortious interference with the plaintiff’s contractual relationship with Twitter. Let’s start with some technical, nominal, cleanup matters. The first named defendant, Scott Gottlieb was both a former FDA Commissioner and a member of Pfizer’s Board of Directors. Maybe this constitutes a disclosure of our priors, but we always found Gottlieb to be one of the more rational commentators when COVID-19 policy was addressed on the Sunday morning shows. The plaintiff also sued President Biden in his official capacity, and that was later automatically changed to President Trump. That is not exactly ironic, but it is a bit strange. Anyway, the plaintiff ended up stipulating to the dismissal of the federal defendants, so the First Amendment claim went away before the Second Circuit could wrestle with it. The Twitter contract was governed by California law, but the interference claim was governed by New York law. Got it? The posture on appeal was that, in a series of orders, the district court dismissed all the claims against everyone. What was on appeal were the dismissals of the section 1985 claim and the New York interference claim. 

The plaintiff’s main beef was that Twitter/X applied its “five strike rule” to suspend the plaintiff’s Twitter account permanently. Under the five strike rule, increasing instances of “Covid-19 misinformation” led to “increasing levels of discipline,” ranging from a warning to temporary account locks to a “permanent suspension.” The five strike rule went into effect on March 1, 2021, while the Covid debate was going full throttle. The plaintiff contacted Twitter about the policy and was told that the points he was raising “should not be an issue at all.” Nevertheless, as the plaintiff tells it, the Biden administration pressured Twitter to misapply the five strikes policy and shut down the plaintiff’s tweets. 

The Second Circuit affirmed the district court’s dismissal of the case. The plaintiff’s §1985 claim failed because antivaxxers and their circle (“Americans who had chosen not to receive a COVID-19 vaccine”) are not a protected class. Rather, such a “class” is of precisely the type that the Supreme Court held does not legally exist − a class of individuals who share a desire to engage in conduct (not getting vaccinated) that the defendants allegedly disfavored.

As an initial matter, there is a difference between a group with immutable characteristics and a group that has certain policy predilections.  The plaintiff tried to get around that distinction by asserting that his proposed affected class included “a disproportionate number of African-Americans, political conservatives, and evangelical Christians.” The Second Circuit expressed skepticism about this forensic maneuver, and swatted it away by holding that section 1985 plaintiffs cannot proceed by arguing disparate impact. More to the point, the court reasoned that the plaintiff’s dispute is that he, personally, was targeted, not any class.  Accordingly, the §1985 claim necessarily failed because the statute is directed against class-based discrimination. 

Next, we get to the claim for tortious interference.  Essentially, the court held that the contract had been enforced by Twitter, not interfered with by the defendants. Twitter’s contract with the plaintiff specified that his account could be terminated at any time for no reason. Thus, according to the court, Twitter’s five-strike policy was actually more generous to plaintiff than necessary under the contract.  None of the plaintiff’s claimed private conversations, with whatever comforting assurances by Twitter personnel, changed that. Thus, there “was no modification of that contract and [the plaintiff] has failed to allege an actual breach of contract.”  

Photo of Michelle Yeary

Here is a question the First Circuit has now answered the wrong way — can a scientific paper published before the FDA approved a drug nonetheless qualify as “newly acquired information” that triggers a duty to change the warnings under the Changes Being Effected regulation? The Fifth Circuit said no last year in Hickey. The Fourth Circuit said no in Knight. The First Circuit’s own prior panel in Celexa said no. And now, in Warner v. Amgen Inc., — F.4th —-, 2026 WL 3033704 (1st Cir. Oct. 9, 2026), the First Circuit has said — maybe yes.

Warner is a hard case. A young man with a history of seizures and cerebrovascular disease died at twenty-five from cerebrovascular complications after taking a single dose of Aimovig, the defendant’s CGRP-receptor antagonist for migraine prevention. Id. at *1. There is no way to write about this one lightly, and we’re not going to try. But the law does not change because the facts are tragic, and the law the First Circuit announced here is, at least in part, a problem.

One quick regulatory note. Aimovig is a biologic, so FDA approval runs through a Biologics License Application (BLA), the biologics counterpart to an NDA. For our purposes the preemption and CBE frameworks are materially the same as on the drug side — there is no express preemption clause for either — and nothing in the analysis below turns on the drug/biologic distinction.

Here is the good news first. Plaintiff originally pleaded a garden-variety pre-approval failure-to-warn claim arguing that the FDA-approved label should have disclosed that clinical trials excluded anyone who had ever had seizures or cerebrovascular disease, and should have warned of the complications that befell her son. Id. at *4. The defendant moved to dismiss on impossibility preemption, pointing out that it was legally required to use the label the FDA approved, and had no unilateral power to change it. Id. The district court agreed, and so did the First Circuit.

The court held, as a matter of law, that a manufacturer’s ability to ask the FDA to include different label language is not the unilateral authority required to defeat an impossibility defense — and that this principle, long established post-approval under Wyeth, Mensing, and Albrecht, applies equally to pre-approval submissions, because a BLA (or NDA) is “by its nature, a request for agency approval.” Id. at *6. That is exactly right, and it is a useful statement of circuit law to have on the books.

The court then added belt to suspenders. The FDA plainly knew the trials excluded people with any history of seizure disorders or significant neurological conditions, and in fact considered and rejected a proposed cerebrovascular warning finding “insufficient evidence to include th[e] theoretical risk in labeling.” Id. at *7. So, the final approved label — with its particular description of the twelve-month exclusion and its silence about the lifetime exclusion — was itself the agency’s “determination” that the label was adequate, and “when the FDA formally approves a label stating one thing with full and obvious notice of the directly contrary position, one can read the approval as rejecting the contrary position.” Id. at *9 (quoting In re Zofran (Ondansetron) Prods. Liab. Litig., 57 F.4th 327, 343 (1st Cir. 2023)).

That is a clean, defense-friendly recitation of the pre-approval preemption rules. We’ll take it. Here is where the opinion starts to drift.

After the hearing on the motion to dismiss, plaintiff changed theories. In a supplemental brief, she submitted nine studies and literature reviews — most of them published before FDA approval — and argued that they were “newly acquired information” that would have allowed the defendant to invoke the Changes Being Effected procedure to unilaterally add a warning after approval but before her son was prescribed the drug. Id. at *5. The district court saw this for what it was and denied leave to amend as futile. The First Circuit reversed. Id. at *17. That alone is a disappointment; but the reasoning is worse.

First, the opinion treats pre-approval publication dates as largely irrelevant. “A study published prior to the approval of a drug or biologic’s label may still qualify as newly acquired information if it had not been ‘previously submitted’ to the FDA.” Id. at *15. Publication dates, the court said, are “not dispositive.” Id.

Compare that with where the rest of the field has landed on the same question. In Hickey v. Hospira, Inc., 102 F.4th 748 (5th Cir. 2024), the Fifth Circuit went the other way. The point there — the point that matters here — is that the pre-approval scientific literature is the baseline. It is what the FDA already had, or was obligated to have, when it made its labeling decision; by definition, it is not “newly” anything. The Fifth Circuit put the proposition in a footnote directly on point, flatly “reject[ing]” the argument “that an individual study that defendant was required to submit to the FDA before approval of Defendants’ drugs independently constitutes ‘newly acquired information.’” Id. at 759 n.8.

The Fourth Circuit got there first, and arguably further. In Knight v. Boehringer Ingelheim Pharmaceuticals, Inc., 984 F.3d 329 (4th Cir. 2021), the court held that information “could not have been ‘newly acquired’ because the FDA already knew about it.” Id. at 340 n.10. The Fourth Circuit specifically concluded that a document the plaintiffs cited “was not ‘newly acquired information’” because “it was submitted to the FDA before it approved Pradaxa.” Id. And even for information not formally submitted, the court held that an association “the FDA already knew of” when it approved the drug — “and nonetheless did not require the defendants specifically to warn of it in the label” — does not qualify as “newly acquired information.” Id. at 338-39. The animating principle is deference. Once something is in the public scientific literature, the agency with the specialized expertise and the statutory charge to evaluate that literature is presumed to have done its job. The question is not whether the defendant physically waved any particular paper in front of the FDA reviewers; the agency — not the manufacturer, and not a court on a motion to dismiss — owns the pre-approval record.

And here is the awkward part for the First Circuit — it had already said the same thing. In In re Celexa & Lexapro Marketing & Sales Practices Litigation, 779 F.3d 34, 41 (1st Cir. 2015) — which the Warner panel itself quotes — the court explained that the “newly acquired information” requirement “lets the FDA be the exclusive judge of safety and efficacy based on information available at the commencement of marketing, while allowing the states to reach contrary conclusions when new information not considered by the FDA develops.” Warner, 2026 WL 3033704, at *10 (quoting Celexa) (emphasis added). It is hard to square “exclusive judge of safety and efficacy based on information available at the commencement of marketing” with a rule that pre-approval publications are “newly acquired” whenever the record is silent on whether the agency saw them.

Second — and this is where it starts to sound like something else by another name — the court repeatedly leaned on the absence of record evidence that the FDA had seen the articles plaintiff relied on. “We also find no evidence in the record as it now stands that the [four studies] were brought to the FDA’s attention.” Id. at *15. Translation — at the pleading stage, we assume the FDA didn’t see this stuff, and the defendant has to come forward later to prove otherwise. That is a backdoor fraud-on-the-FDA theory, only without the discipline Buckman Co. v. Plaintiffs’ Legal Committee, 531 U.S. 341 (2001), would impose. Buckman, notably, is nowhere in the opinion. And that is the whole point of Buckman. Deciding whether the FDA was properly informed — whether it was fully in the loop or somehow hoodwinked — is the FDA’s job, not a federal court’s. Treating a plaintiff’s bare assertion that studies were not “brought to the FDA’s attention” as enough to defeat preemption at the pleading stage runs headlong into that principle.

Third, the First Circuit opened the door to animal studies as “newly acquired information.” The court was “not certain that the [animal] study fails to qualify as newly acquired information,” reasoning that FDA guidance permits “theoretical warnings ‘if the animal data raises substantial concern about the potential for … adverse reaction in humans.’” Id. at *16. We understand the hook — Aimovig was a first-in-class biologic and the FDA itself relied on animal data in approving it. Id. But using that fact to justify forcing manufacturers to chase every rodent paper into the CBE process is going to produce exactly the overwarning the agency keeps telling us it doesn’t want. The court even acknowledged earlier in the opinion that the FDA aims to “prevent overwarning,” including “theoretical hazards not well-grounded in scientific evidence [that] can cause meaningful risk information to lose its significance.” Id. at *2 (citations omitted). The two halves of the opinion do not entirely talk to each other.

Fourth, the causal-association prong took a hit. The court held that the regulatory definition of an adverse reaction does not “require[] that the inclusion of an adverse reaction in labeling be based solely on clinical studies of the drug itself,” id. at *18, and credited plaintiff’s theory that the newly acquired studies showed Aimovig was associated not with causing seizures, but with worsening the outcomes of seizures by weakening the blood-brain barrier. Id. at *17. That is a notably loose formulation of causal association for a motion-to-dismiss standard. The CBE regulation’s two causal thresholds — “reasonable evidence of a causal association” for warnings and precautions, and “some basis to believe there is a causal relationship” for adverse reactions, id. at *17 — were already generous. They are now even more so in the First Circuit.

Finally, we note what the opinion does not decide. The court flagged the “clear evidence” defense — that even if the four surviving studies had been submitted, the FDA would still have rejected the proposed label change — but observed that the defendant had “chosen not to rely on such an argument” at this stage, so the court didn’t reach it. Id. at *19. That defense remains fully available on remand, and given how specifically the FDA engaged with CGRP antagonism and the “world’s literature” on the subject, we would expect it to feature prominently on summary judgment.

The scorecard isn’t exactly even. Dismissal of the original pre-approval theory affirmed. Good. Clear circuit-level statement that a BLA/NDA submission is a request for agency action and cannot defeat impossibility preemption. Better. A ruling that pre-approval studies can be “newly acquired information,” that single animal studies may qualify, and that the required causal association does not need to involve the drug itself. Not good. And the whole thing is wrapped in a pleading-stage “we cannot say with certitude” posture that puts the manufacturer in the position of having to disprove, later, what plaintiffs barely alleged in the first place.

The bright spot is that this is a remand, not a verdict. The court itself noted that it was “inclined to read the record as the district court did” on the newly-acquired-information issue, and reversed only because, without expert testimony, it could not get all the way to “certitude.” Id. at *14. Plaintiff gets to replead. Based on what she has put on the table so far, we would not bet the farm on where she ends up.

But the broader doctrinal footprint is the thing to watch. The Fourth Circuit in Knight and the Fifth Circuit in Hickey — and, before them, the First Circuit’s own panel in Celexa — had put the field on reasonably firm ground that information already in the FDA’s hands (or in the public scientific literature) at the time of approval is not “newly acquired” when a plaintiff later rediscovers it. Warner nudges that ground. If a published-but-not-formally-submitted study predating approval can still ground a CBE claim, “the exclusive judge of safety and efficacy” is no longer the agency — it is the lawyer with the better literature search. We hope the First Circuit, or the next panel to grapple with this on a fuller record, reels this one back in.

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