Part of the AI Regulation News hub.
A Delaware Court of Chancery report treats fictitious quotations from real cases as a hallmark of AI usage, and calibrates culpability by distinguishing a disbarred attorney from the self-represented litigant seeking access to the court
The detection point is precise: not a fabricated case, but real cases quoted with words they do not contain. And in one sentence the court sets out a sliding scale of culpability that most AI-in-filings decisions leave implicit.
Bottom line: A final report under Court of Chancery Rule 144 in a trust dispute. The Master found that the self-represented respondent had twice quoted a real decision using words that never appear in it, described fictitious quotations as one of the hallmarks of AI usage, and recorded that abuse of AI in litigation filings is sanctionable conduct. The citation conduct formed part of a finding of subjective bad faith, and fees and costs were shifted to the petitioners.
Who this affects: Anyone filing in Delaware, and more generally anyone assessing how courts detect AI-assisted filings. The detection signal here is different from the fabricated-citation cases and harder to catch.
Issue date: Report dated 10 August 2026; matter submitted 23 April 2026.
What changed: No rule changed. The report contributes a detection criterion, an explicit link to Delaware's existing AI hallucination authority, and a statement calibrating culpability to the filer's sophistication.
Analysis: Note what was actually wrong with the citations. The cases were real and the reporter citations were real. What was fabricated was the quoted language attributed to them. That defeats the standard verification habit of checking whether a case exists, and it is the reason the court reached for the word hallmark rather than treating it as an ordinary citation error.
Primary sources: Palumbo v. Palumbo, report of 10 August 2026, Delaware Courts · Case record, CourtListener
- Instrument
- Palumbo v. Palumbo, Master's report in the Court of Chancery
- Citation
- C.A. No. 2024-0661-DH
- Court
- Court of Chancery of the State of Delaware, Master Hume IV
- Parties
- Stephen J. Palumbo and Lori A. Palumbo as co-trustees, petitioners; Gregory M. Palumbo, respondent, self-represented
- Jurisdiction
- Delaware, United States
- Status
- Report dated 10 August 2026.
- Status of the report
- The document states that it is a final report pursuant to Court of Chancery Rule 144. A final report is the Master's concluded recommendation, and it remains subject to the exceptions process and review by a judge of the Court of Chancery, so it is not itself a final judgment of the court.
- Outcome
- The Master recommended that the petitioners' motion to remove a trustee be granted, that the respondent's cross-petition for an accounting be denied, and that fees and costs be awarded to the petitioners under the bad faith exception to the American Rule.
- Authority relied on
- An v. Archblock, 2025 WL 1024661 (Del. Ch. Apr. 4, 2025), which defines AI hallucination and characterises abuse of AI in filings as an abuse of the adversary system
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://courts.delaware.gov/opinions/download.aspx?id=399480
The detection point is quotation, not existence
The respondent's post-trial briefing cited Kuroda on two occasions. The court found that the cited words never appear in the Kuroda decision.
The footnote is specific about both instances. One quoted sentence about dismissal not being warranted where the defendant is on notice of the claim carried emphasis in the respondent's brief but, as the court noted, there was no emphasis in the original because the language does not appear there at all. A second proposition, that the court looks to the substance of the allegations rather than the form, was one the court said Kuroda never says nor suggests.
A second authority fared no better: the court held that Koninklijke v. Philips Electronics does not stand for the proposition that a Rule 12(e) motion is the appropriate response to unclear pleading that mixes theories.
This is a materially different failure from the fabricated-case line of decisions. Here the cases exist and the citations resolve. What does not exist is the quoted language. A verification step that confirms a case is real and correctly cited will pass this filing, which is precisely why the court treated the pattern as diagnostic.
What the court said it indicates
The court stated that inclusion of fictitious quotations from cases constitutes one of the hallmarks of Artificial Intelligence usage.
It grounded that in existing Delaware authority rather than asserting it fresh, citing An v. Archblock for the definition of AI hallucination as a phenomenon in which a large language model perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate.
From the same authority it drew the consequence: abuse of AI in litigation filings comprises an abuse of the adversary system and is sanctionable conduct.
The word hallmark should be read carefully. The court did not find as a fact that AI produced the briefs, and it did not need to. It identified fictitious quotation as an indicator, then proceeded on the basis of what the filings actually contained, which was false authority regardless of how it was generated.
The sentence that calibrates culpability
The most quotable line in the report is a distinction: the court observed that the respondent is not the typical self-represented litigant seeking to use artificial intelligence to obtain access to the Court, and that he is a seasoned attorney who understands the repercussions of submitting false authority to a court.
Two things are embedded in that. The first is a description of the ordinary case that is notably sympathetic. The court's working assumption is that a self-represented litigant reaching for AI is trying to obtain access to the court, which frames the typical pro se AI user as someone attempting to participate rather than someone attempting to deceive.
The second is that culpability is calibrated to the filer's sophistication. The respondent, described in the report as a disbarred attorney appearing pro se, was treated as someone who knew what submitting false authority meant. His formal status was self-represented; his actual knowledge was that of a lawyer, and the court measured him by the latter.
Read alongside the Pennsylvania Superior Court's decision in Kaspryak, which extended the same concern to pro se litigants but disposed of the appeal by waiver rather than sanction, a pattern is visible. Courts are separating the consequence that follows automatically from defective briefing from the sanction that follows from culpable conduct, and locating the filer on a scale between the two.
The consequence the citation conduct contributed to
The report should not be read as one in which nothing followed. The Master made a finding of subjective bad faith and shifted fees, and the citation conduct is named as part of what supported it.
Setting out what he called a concert of action, the Master included that the respondent committed critical errors in citing law in his post-trial brief, relying on Delaware cases for propositions neither supported nor implicated in the case. Alongside it he listed the use of cross-examination to press witnesses about matters ancillary or irrelevant to the petitions, a misrepresentation about the scheduling of post-trial briefs contradicted by documentary proof, and a failure to make any case in support of the cross-petition which imposed undue burdens on the petitioners.
His conclusion was that this concert of action supports the finding that the respondent conducted the litigation with subjective bad faith, and that as a result he shifts fees and costs to the petitioners.
So defective citation was not treated as a standalone offence attracting its own penalty. It was one strand of conduct feeding a fee-shifting finding under the bad faith exception to the American Rule, which is a different route from a sanction aimed at the filing itself, and for a client a more expensive one.
What to take from it
For verification practice, the operative lesson is that confirming a case exists is not enough. The quoted language has to be checked against the decision, because the failure mode here survives every check short of reading the source.
For anyone assessing exposure, the Delaware position is now stated in two places. An v. Archblock supplies the definition and the characterisation of abuse of AI in filings as an abuse of the adversary system and sanctionable conduct, and this report applies it.
One caveat on status, which matters for citation. This is a Master's report, a recommended disposition subject to exceptions and review by the Court of Chancery, not a final judgment. Its reasoning is useful and its authority is limited, and it should be cited for what it is.
The wider point for institutions writing AI policies is the detection criterion. A policy that instructs staff to verify that cited authorities exist addresses the previous generation of this problem. A policy that requires the quoted words to be located in the source addresses this one.
Three things are worth extracting. First, the detection criterion: the Master treated fictitious quotations from real cases as one of the hallmarks of AI usage, a harder failure than a fabricated case because the citation resolves and only the quoted words are invented. Any verification rule that stops at confirming a case exists will not catch it, and the fix is to require that quoted language be located in the source. Second, the calibration: he distinguished this respondent from the typical self-represented litigant seeking to use artificial intelligence to obtain access to the court, on the ground that he is a seasoned attorney who understands the repercussions of submitting false authority. Third, and easily missed: something did follow. Critical errors in citing law formed part of the concert of action supporting a finding of subjective bad faith, and fees and costs were shifted to the petitioners under the bad faith exception to the American Rule. Note the status: this is a final report under Court of Chancery Rule 144, still subject to exceptions and review by a judge, so not itself a final judgment of the court.
Source File
https://courts.delaware.gov/opinions/download.aspx?id=399480
Open the report PDF and confirm four things: the caption showing the Court of Chancery of the State of Delaware, C.A. No. 2024-0661-DH, with the report dated 10 August 2026 and Gregory M. Palumbo appearing pro se; the passage stating that the respondent purports to cite Kuroda on two occasions yet the cited words never appear in the Kuroda decision; footnote 106, stating that inclusion of fictitious quotations from cases constitutes one of the hallmarks of Artificial Intelligence usage and citing An v. Archblock; and the sentence immediately following the Koninklijke discussion, distinguishing the respondent from the typical self-represented litigant seeking to use artificial intelligence to obtain access to the Court.
Inclusion of fictitious quotations from cases constitutes one of the hallmarks of Artificial Intelligence usage. Court of Chancery of the State of Delaware, Palumbo v. Palumbo, C.A. No. 2024-0661-DH, report of 10 August 2026, footnote 106
FAQ
What exactly was wrong with the citations?
The cases were real but the quoted language was not in them. The court found that the respondent cited Kuroda twice using words that never appear in that decision, and that a second authority did not stand for the proposition attributed to it.
Did the court find that AI was used?
It did not make that finding. It stated that fictitious quotations are one of the hallmarks of AI usage, and then proceeded on the basis of what the filings contained, namely false authority, without needing to determine how it was produced.
What does Delaware say about AI in filings?
The report relies on An v. Archblock, which defines AI hallucination and states that abuse of AI in litigation filings comprises an abuse of the adversary system and is sanctionable conduct.
Why does the court mention self-represented litigants?
To distinguish this respondent. It observed that he is not the typical self-represented litigant seeking to use artificial intelligence to obtain access to the Court, but a seasoned attorney who understands the repercussions of submitting false authority. The report records that he is a disbarred attorney appearing pro se.
Is this binding authority?
It is a final report pursuant to Court of Chancery Rule 144, the Master's concluded recommendation. It remains subject to the exceptions process and review by a judge of the Court of Chancery, so it should not be cited as a final judgment of the court.
Did anything follow from the citation conduct?
Yes. The Master listed the respondent's critical errors in citing law in his post-trial brief among the conduct making up a concert of action supporting a finding of subjective bad faith, and shifted fees and costs to the petitioners under the bad faith exception to the American Rule. It was not treated as a standalone offence with its own penalty.
Why is this failure harder to detect than a fabricated case?
Because the case exists and the citation resolves. Verification that stops at confirming the authority is real will pass the filing. Catching it requires locating the quoted words in the decision itself.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.