Supreme Court Quashes ₹425-Crore Customs Penalty Built on AI-Hallucinated Case Law

A diamond trader imported a consignment. Customs said the diamonds were natural, mis-declared as lab-grown to dodge a higher tariff, and slapped a penalty of ₹425.27 crore — a figure large enough to end most businesses outright. The trader fought it, lost at the Gujarat High Court, and took it to the Supreme Court. What undid the penalty there wasn’t a dispute about diamonds at all. It was what the adjudicating officer had cited to justify it.

The case is Vijay Ghanshyam Gadiya v. Union of India and Another [2026 LiveLaw (SC) 884], decided on 2 September 2026 by a Bench of Justice Dipankar Datta and Justice Sheel Nagu. The penalty had originally been imposed by the Additional Commissioner of Customs, Surat, on 8 October 2025, under Section 114 of the Customs Act, 1962, for allegedly mis-declaring natural diamonds as lab-grown ones to pay a lower rate of duty. The Gujarat High Court dismissed the trader’s challenge on 20 January 2026. Before the Supreme Court, the trader’s counsel raised a specific, unusual objection: several of the judgments and articles the Additional Commissioner had cited in the original order, he said, appeared to have been generated by artificial intelligence.

Sidebar: the Bench didn’t take the allegation on faith — it went and checked. That step matters as much as the finding itself. A court independently verifying whether cited precedent actually exists, rather than assuming a departmental order’s citations are accurate, is itself a new layer of scrutiny that adjudicating officers should expect going forward, in tax matters as much as customs.

The Bench said it independently verified the cited material and found that some of the cases relied upon were non-existent or carried fake citations altogether. A further check revealed that other cases did exist — but did not lay down the legal propositions the officer had attributed to them, which the Court described as an apparent “hallucination of AI.” In its own words: “Upon verifying such judgments and articles, it does appear that the second respondent has relied upon case laws that are either non-existent or have fake citations. A further investigation has also revealed that some of the case laws relied upon by the second respondent, which do exist, do not lay down the ratio deduced therefrom and it appears to be a hallucination of AI.”

The Court referred to its own recent decision in Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. [2026 LiveLaw (SC) 653], which had already established a zero-tolerance approach to unverified AI-generated legal material — holding that an advocate citing such material without verification could face misconduct proceedings, and that judicial or quasi-judicial reliance on fake or hallucinated precedent is a serious lapse going to the integrity of adjudication itself. The Bench noted that the Supreme Court has published draft Regulations for the Use of Artificial Intelligence in Courts, and that even though those regulations are not yet finalised, the principle they embody already applies: “assistance can never be substituted for adjudication. AI may well serve as training wheels but entrusting it with the pilot’s seat would be both imprudent and dangerous.” Holding that reliance on this dubious material was fatal to the sustainability of the penalty order, the Court set aside both the Gujarat High Court’s judgment and the original Order-in-Original, revived the proceedings for fresh adjudication by an officer of the same rank — but not the officer who passed the original order — and left it open to the appointing authority to consider action against the author of the original order.

Why It Matters

This is Finoscape’s third article on AI-hallucinated legal material undoing a tax or regulatory order in 2026 — after an earlier Supreme Court ruling that set aside an NCLT order on the same ground, and the Gujarat High Court’s Faiz Enterprise ruling quashing a GST cancellation order for the identical reason. Three independent instances, across three different fora (NCLT, a State GST authority, and now a Central Customs adjudicating authority) and within a matter of months, is no longer an isolated curiosity — it is a pattern serious enough that the Supreme Court itself is now treating it as a standing area of institutional concern, complete with draft judicial AI regulations already in circulation. For CA firms, the direct relevance cuts two ways: first, as advisors, this reinforces a fresh, high-authority citation to challenge any departmental order that relies on case law the firm cannot independently locate or verify; second, and just as importantly, it is a direct caution to the profession’s own growing use of AI-assisted legal and tax research tools, since the same “verify before you rely” standard the Court is now applying to adjudicating officers applies with equal force to any advisor citing AI-suggested precedent in a submission, appeal, or client opinion.

Key Takeaways

• The Supreme Court quashed a ₹425.27 crore Customs penalty under Section 114 of the Customs Act, 1962, after independently verifying that several case laws relied upon by the adjudicating officer were non-existent, carried fake citations, or did not support the propositions attributed to them — an apparent AI hallucination.
• This is the third such instance in 2026 across three different adjudicatory/tribunal fora that Finoscape has now covered (NCLT, Gujarat HC/GST, and now the Supreme Court/Customs), indicating a genuine, recurring pattern rather than an isolated lapse.
• The Court reaffirmed the zero-tolerance standard set in Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. [2026 LiveLaw (SC) 653]: citing unverified AI-generated legal material can constitute advocate misconduct, and judicial/quasi-judicial reliance on such material is a serious lapse affecting the integrity of adjudication.
• The Court’s own formulation — “AI may well serve as training wheels but entrusting it with the pilot’s seat would be both imprudent and dangerous” — squarely permits AI as an assistive research tool while barring it from substituting for actual adjudicative reasoning.
• The matter was revived for fresh adjudication by a different officer of the same rank, with the appointing authority left to consider whether action should be taken against the original order’s author — signalling that reliance on hallucinated material carries potential personal, not just institutional, consequences for the officer concerned.

Practical Implications

Firms defending any pending tax, GST or Customs proceeding should now routinely ask for, and independently verify, the full text and citation of every judgment relied upon in a departmental order or show cause notice — this ruling, following Faiz Enterprise and the earlier NCLT ruling, gives firms a well-established, high-authority basis to challenge an order on exactly this ground wherever a citation cannot be located or does not say what the order claims it says. Equally, firms that have begun using AI tools for legal research, drafting submissions, or preparing opinions should treat this ruling as a direct, practice-level warning: every AI-suggested case citation must be independently verified against a primary source before it is used in any client-facing document, appeal, or representation, given the Court’s explicit endorsement of professional-misconduct consequences for unverified reliance. Firms should also consider a standing internal verification checklist for AI-assisted legal research specifically, given how frequently this exact fact pattern is now recurring across different tribunals and courts.

Action Checklist

• Build a standing practice of independently verifying every case citation in any departmental order, show cause notice, or adjudication before responding to or appealing it — do not assume departmental citations are accurate.
• Where a cited case cannot be located, or exists but does not support the proposition attributed to it, raise this specifically and explicitly as a ground of challenge, citing Vijay Ghanshyam Gadiya, Faiz Enterprise, and the Court’s own NCLT ruling as a developing line of authority.
• If the firm uses AI tools for legal or tax research, institute a mandatory verification step for every AI-suggested citation before it appears in any submission, appeal, or client opinion — treat this as a professional-risk control, not an optional best practice.
• Track the Supreme Court’s draft Regulations for Use of Artificial Intelligence in Courts as they are finalised, since they are likely to shape both judicial and departmental practice going forward.
• Brief relevant client-facing teams on this ruling specifically where the firm advises on Customs, GST, or NCLT matters, given how directly it bears on active adjudications in those fora.

Relevant Sections / Rules / Notifications

• Section 114, Customs Act, 1962 (penalty provision under which the original order was passed)
• Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. [2026 LiveLaw (SC) 653] (zero-tolerance standard on unverified AI-generated legal material; advocate misconduct and adjudicatory integrity)
• Supreme Court’s draft Regulations for Use of Artificial Intelligence in Courts (referenced by the Bench; not yet finalised)
• Finoscape’s own prior coverage: the Gujarat High Court’s Faiz Enterprise v. State Tax Officer ruling (AI-generated case laws in a GST cancellation order) and an earlier Supreme Court ruling setting aside an NCLT order on AI-hallucinated citations

FAQs

Q: Does this ruling ban the use of AI in tax or legal research altogether?
A: No. The Court explicitly said AI can serve as an assistive tool to speed up decision-making, and did not prohibit its use. What it barred is unverified reliance — using AI-generated material, including case citations, without independently confirming it is accurate before it is relied upon in an order, submission, or opinion.

Q: Is this the first time an Indian court has set aside a tax-related order for AI-hallucinated case law?
A: No — it is the third such instance Finoscape has now covered in 2026, after an earlier Supreme Court ruling on an NCLT order and the Gujarat High Court’s Faiz Enterprise ruling on a GST cancellation order. This is a developing pattern, not a one-off.

Q: What happens to the underlying Customs dispute now?
A: The Supreme Court did not decide the merits of the mis-declaration allegation itself. It set aside both the Gujarat High Court’s judgment and the original Order-in-Original on the AI-hallucination ground, and directed fresh adjudication by a different officer of the same rank — meaning the underlying dispute over classification of the diamonds remains open and will be reconsidered afresh.

Internal Links

Today’s Intelligence — 8 September 2026 · Case Law hub

Related Articles

Gujarat High Court — AI-Generated Case Laws in a GST Order: Gujarat HC Quashes Proceedings, Issues Verification Protocol, Warns of Contempt (Finoscape, prior coverage) · Supreme Court — Supreme Court Sets Aside NCLT Order Over AI-Hallucinated Case Citations (Finoscape, prior coverage)

Prepared by Finoscape Editorial Team — hello@finoscape.com. This article is for general informational purposes and does not constitute legal or professional advice, and is based on LiveLaw’s and Bar and Bench’s byline-attributed reporting of the Supreme Court’s judgment, rather than direct retrieval of the order from the Supreme Court’s own website. Practitioners should independently verify the order before citing it in client advice or submissions.