A Washington federal court excluded an expert report for containing AI-hallucinated citations, then granted summary judgment and dismissed the case with prejudice

What the order actually does

The court granted three motions in one document. It excluded a regulatory expert, Ms. Evelyn Cadman (Dkt. 206), because her opinion offered legal conclusions rather than specialized knowledge. It excluded a psychiatric expert, Dr. Gabriel Holguin (Dkt. 208), on reliability grounds. It then granted the defendants' motion for summary judgment (Dkt. 207) and dismissed the plaintiff's remaining claims "WITH PREJUDICE and without leave to amend."

The plaintiff, an Army nurse, alleged that nootropic supplement kits were adulterated with amphetamines and undisclosed ingredients. A January 2026 partial summary judgment order had already knocked out her primary theory, that the supplements caused a positive drug screening. What remained was a claim that adulterated products caused negative health effects, plus a claim for the purchase price she paid.

Both remaining routes closed. The medical damages claim failed for want of admissible expert causation evidence. The purchase price claim was mooted by an unconditional tender of $13,000, which exceeded the $12,708.78 the defendants calculated as the ceiling on a $632 purchase.

The citation table, and how it reached two experts

The AI problem in this case did not start with Dr. Holguin. The court had already issued sua sponte sanctions against plaintiff's attorney, Ms. Jocelyn Stewart, at Dkt. 265, for what this order describes as "generating numerous false citations with artificial intelligence and submitting them to the Court without verification."

The mechanism is stated plainly, and it is the part experts should read twice. Quoting its own sanctions order, the court records that the plaintiff admitted "she used Claude or ChatGPT to 'generate a formatted citation table' for academic articles and 'provided the same AI-generated citation table to both experts as an appendix, and neither expert caught the errors in the citation data before signing their reports.'"

So one artifact, a formatted appendix of academic references, was produced by a chatbot in counsel's office and then attached to the reports of two separate experts, Mr. James Kababick and Dr. Ronald Shippee. Neither checked it. Both signed.

Dr. Holguin's report was a third carrier. The court notes that plaintiff's counsel was never asked to explain those particular citations, but found the errors "similar to those in other expert reports" produced the same way.

Why withdrawing the tainted pages did not save the report

This is the finding with the longest reach. The plaintiff had already withdrawn three of the four opinions in Dr. Holguin's report, and the hallucinated citations lived in the withdrawn portions. The surviving fourth opinion, on a symptom course consistent with stimulant withdrawal, carried none of them.

The court excluded the report anyway. In its words: "Thus, although these errors occur only in the withdrawn portions of Dr. Holguin's opinion, they 'undermine[] the overall credibility' of the report." It then stated the holding in one line: "The Court finds that the hallucinated citations undermine Dr. Holguin's entire report and that the report is not reliable under Daubert."

The reasoning is squarely Rule 702. The court found that counsel's level of involvement in drafting the reliance list with AI, combined with "Dr. Holguin's lack of diligence in verifying the sources that he represented were the basis for his opinion," prevented the plaintiff from carrying her burden under Fed. R. Evid. 702(b) through (d), the sufficient-facts, reliable-methods and reliable-application prongs.

Two borrowed lines did the rhetorical work. The court quoted Kohls v. Ellison for the proposition that the hallucinated citations "shatter[] his credibility with this Court," and Concord Music Group, Inc. v. Anthropic PBC for the overall credibility point. Our reading, and it is our reading rather than the court's: the judge treated the reliance list as an inseparable part of the report's foundation, so a defect anywhere in it went to the whole.

The sanction was not a fine, it was the case

Trace the causal chain in the order itself. Exclude Dr. Holguin. Then, at Section IV.C.1, the court finds that the plaintiff's medical damages theory is "medically complex" under Washington law and "would require expert testimony at trial," and that "no remaining expert in the record discusses how the supplements could have caused withdrawal symptoms or any other medical damages."

The court added a belt-and-braces holding. Even considering Dr. Holguin's opinion despite its unreliability, it "does not speak directly to medical causation," because he "indisputably never opined on more-likely-than-not causation." So the exclusion was not strictly the only reason the claim died. But it was the first reason, and it was the one the court reached first.

That matters for how the AI-citation problem gets classified. Until now the visible consequences of fabricated citations have mostly been professional: fines, show-cause orders, referrals, bar discipline. Here the consequence landed on the merits. The client lost the case with prejudice.

For a testifying expert, the practical translation is short. A reliance list you did not personally check is not a formatting detail. On this record, it was enough for a judge to find your entire report unreliable, including the parts nobody complained about.

An apparent split, which we did not confirm

There is reportedly a contrary order six days earlier. In re: Bard Implanted Port Catheter Products Liability Litigation, an MDL in the District of Arizona, is reported to be dated 2026-08-12 and to have declined to exclude an expert on AI-hallucination grounds.

We did not open that order. We are flagging it because a reader who takes LeDoux as settled law would be building on one side of what may be a live disagreement, and because a district court order on either side settles nothing beyond its own docket.

State the position honestly and it is narrow: one judge in the Western District of Washington, on this record, found that hallucinated citations in an expert's reliance list destroyed the report's reliability under Rule 702. That is not "federal courts now require" anything. It is not a holding that AI citations automatically invalidate an expert report. It is not authority that counsel may never use AI. Any of those framings would overstate a single unappealed district court order.

What a report preparer should change on Monday

The duty that transfers is narrow and testable. An expert should personally open and confirm every source in the reliance list of a report bearing their signature, and counsel should not hand an expert a citation appendix generated by a chatbot.

The failure mode in LeDoux was not exotic. It was a convenience: counsel formatted a reference table with a language model, sent the same file to two experts, and nobody treated the appendix as substantive content. The file spread across reports precisely because it looked clerical.

Worth noting what the court did not do. It did not sanction Dr. Holguin, and it did not suggest a curative process, a supplemental declaration or a corrected appendix, that would have rescued the report. The plaintiff had tried a clarifying declaration; the court declined to consider it because it arrived more than four months after the expert disclosure deadline.

What we did not verify

What we opened: the full 19-page order, Dkt. 269, filed 08/18/26 in Case No. 3:24-cv-05808-TMC, in PDF, from the link in the primary sources above. Both quotations reported to us matched the extracted text character for character, including the nested single quotes in the citation-table admission. The date is taken from the instrument itself, from the signature block and the ECF header, not from any trade report.

What we did NOT open: the sanctions order against plaintiff's counsel (Dkt. 265), the expert reports themselves (Dkts. 206-1, 208-1, 208-2), the motions (Dkts. 206, 207, 208), the January 2026 partial summary judgment order (Dkt. 202), the docket, and any subsequent judgment or notice of appeal. We also did NOT open the In re: Bard Implanted Port Catheter order reported to be dated 2026-08-12, so we cannot describe what it held, how similar its facts were, or whether a genuine split exists. Everything above about Bard is unconfirmed report, not verified content.

What we refuse to claim: that this reflects a trend, a majority position, or an emerging federal standard; that other courts will follow it; that any specific number of citations was hallucinated, since the order gives no count; that Dr. Holguin or the other experts used AI themselves, which the order does not say; that the case is over, since we did not check for an appeal; and that AI-assisted citation work is per se disqualifying, which this order does not hold.

Related briefings

• Expert AI prompts held discoverable under Rule 26
• Proposed Federal Rule of Evidence 707 on AI-generated evidence
• Eleventh Circuit sanctions over AI citations
• California bar guidance on verifying AI output
• AI Regulation News hub