What was hallucinated

Misrepresented: Exhibits & Submissions | Party submitted ChatGPT-derived statements; Court disregarded them because the underlying question/prompt was unknown, so the content lacked reliable provenance. || Misrepresented: Case Law | Party made a general reference to judgments of the Arnhem-Leeuwarden Court of Appeal without providing sources; Court disregarded the claim.

Details

AI UseThe appellant relied on ChatGPT to generate a list of ten "economically comparable" vehicles for purposes of arguing a lower trade-in value to reduce bpm (car registration tax). The Court noted this explicitly and criticized the mechanical reliance on AI outputs without human verification or contextual adjustment.Hallucination DetailsChatGPT produced a list of luxury and exotic cars supposedly comparable to a Ferrari 812 Superfast. The Court found that mere AI-generated association of vehicles based on "economic context and competition position" is insufficient under EU law principles requiring real-world comparability from the perspective of an average consumer.Ruling/SanctionThe Court rejected the appellant’s valuation arguments wholesale. It stressed that serious, human-verified reference vehicle comparisons were mandatory and that ChatGPT lists could not establish the legally required comparability standard under Dutch and EU law (Art. 110 TFEU). No monetary sanction imposed, but appellant’s entire case collapsed on evidentiary grounds.Key Judicial ReasoningThe Court reasoned that a list generated by an AI program like ChatGPT, without rigorous control or verification, is inadmissible for evidentiary purposes. AI outputs lack the nuanced judgment necessary to assess "similar vehicles" under Art. 110 TFEU and Dutch bpm tax rules. It underscored that the test is based on the perceptions of a human average consumer, not algorithmic proximity.

Sanction teardown · The Hague CA, Netherlands · 2024-06-26

X BV in Z v. Tax Inspector

What happened

In The Hague CA, Netherlands, a filing relied on ChatGPT to help draft legal argument. The court identified the following problems with the citations in that filing:

Misrepresented (Exhibits & Submissions)

Party submitted ChatGPT-derived statements; Court disregarded them because the underlying question/prompt was unknown, so the content lacked reliable provenance.
Misrepresented (Case Law)

Party made a general reference to judgments of the Arnhem-Leeuwarden Court of Appeal without providing sources; Court disregarded the claim.

Which AI tool

ChatGPT. Note: Charlotin's public database records tool attribution only where a court order, brief, or reporting on the matter states it explicitly; "unidentified" or "implied" means the record indicates AI use but does not name a specific product — we do not guess.

Outcome

Arguments rejected; No formal sanction but severe judicial criticism.

Additional detail

AI UseThe appellant relied on ChatGPT to generate a list of ten "economically comparable" vehicles for purposes of arguing a lower trade-in value to reduce bpm (car registration tax). The Court noted this explicitly and criticized the mechanical reliance on AI outputs without human verification or contextual adjustment.Hallucination DetailsChatGPT produced a list of luxury and exotic cars supposedly comparable to a Ferrari 812 Superfast. The Court found that mere AI-generated association of vehicles based on "economic context and competition position" is insufficient under EU law principles requiring real-world comparability from the perspective of an average consumer.Ruling/SanctionThe Court rejected the appellant’s valuation arguments wholesale. It stressed that serious, human-verified reference vehicle comparisons were mandatory and that ChatGPT lists could not establish the legally required comparability standard under Dutch and EU law (Art. 110 TFEU). No monetary sanction imposed, but appellant’s entire case collapsed on evidentiary grounds.Key Judicial ReasoningThe Court reasoned that a list generated by an AI program like ChatGPT, without rigorous control or verification, is inadmissible for evidentiary purposes. AI outputs lack the nuanced judgment necessary to assess "similar vehicles" under Art. 110 TFEU and Dutch bpm tax rules. It underscored that the test is based on the perceptions of a human average consumer, not algorithmic proximity.

How Citation Safe would have caught this

Citation Safe runs three deterministic layers before a brief is filed: (1) does the citation exist against CourtListener's database of published opinions, (2) if quoted, does that exact language appear in the source, (3) does the cited case actually support the proposition it is cited for. Fabricated case citations fail Layer 1. Fabricated or misattributed quotations fail Layer 2 even when the underlying case is real. Misrepresented holdings — a real case cited for a proposition it does not support — are the target of Layer 3. None of these checks involve asking another language model whether the citation looks right; they are lookups and text-matches against the actual source, which is why a hallucinated citation has to survive a direct lookup against the authoritative source — not another model's opinion — to earn a VERIFIED stamp; our measured false-verify rate is published live at /quality.

Check a brief before you file it → · See our live false-verify rate

Source: https://www.damiencharlotin.com/documents/258/X_BV_in_Z_v._Tax_Inspector_Hague_Court_of_Appeal_June_26_2024.pdf, via Damien Charlotin's public AI Hallucination Cases Database (CC0).