Supreme Court sets aside ₹425 Cr Customs Penalty over fake, AI Hallucinated citations

The Supreme Court has set aside a customs penalty of over ₹425 crore after finding that the adjudicating authority relied on non-existent judgments, fake citations, and incorrect propositions attributed to genuine judgments, which appeared to be AI-generated hallucinations.

Case Details

Case: Vijay Ghanshyam Gadiya v. Union of India and Another
Court: Supreme Court of India
Bench: Justice Dipankar Dutta and Justice Sheel Nagu
Date: September 2, 2026
Legislation: Customs Act, 1962
Provision: Section 114

Background

Vijay Ghanshyam Gadiya was issued an Order-in-Original dated October 8, 2025, by the Additional Commissioner of Customs, Surat.

The adjudicating officer imposed a penalty of ₹425,27,99,100 under Section 114 of the Customs Act, 1962, alleging that a consignment of natural diamonds had been mis-declared as lab-grown diamonds with the intention of paying a lower tariff.

Gadiya challenged the order before the Gujarat High Court. The High Court dismissed his appeal on January 20, 2026.

He thereafter approached the Supreme Court.

Supreme Court Examines AI-Generated Legal Material

Before the Supreme Court, Gadiya contended that the adjudicating authority had relied upon judgments and legal articles generated using artificial intelligence.

The Bench stated that it had “made the effort to individually verify whether such contention of the appellant is true.”

The Court’s verification found that the adjudicating authority had relied upon case laws that were either non-existent or carried fake citations.

It further found that while some of the cited cases did exist, they did not lay down the legal propositions attributed to them.

The Court observed:

“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.”

The Bench further observed:

“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.”

Reliance on Pooja Ramesh Singh

The Supreme Court referred to its earlier decision in Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., which dealt with the use of artificial intelligence in judgments.

The Court reiterated the need for verification of AI-generated legal authorities and referred to the earlier observation:

“It is necessary for courts to adopt a zero-tolerance mode for producing, citing or using artificial intelligence-generated precedents without verification.”

The earlier judgment had also held that citing AI-generated judgments without verification constitutes misconduct on the part of an advocate.

“AI May Well Serve as Training Wheels”

The Bench also considered the broader role of AI in adjudication, particularly in the context of the Supreme Court’s recently released draft Regulations for Use of Artificial Intelligence in Courts, 2026.

While recognising that AI can assist in expediting decision-making, the Court emphasised that such assistance cannot replace the adjudicatory function.

The Bench observed:

“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.”

Supreme Court’s Decision

The Supreme Court held that the reliance on the dubious AI-generated material was “fatal to the sustainability of the order imposing penalty on the appellant.”

The Court accordingly:

• Set aside the Gujarat High Court’s order dated January 20, 2026;
• Set aside the Order-in-Original dated October 8, 2025, imposing the ₹425,27,99,100 penalty;
• Revived the proceedings for fresh adjudication by a different officer of the same rank; and
• Left it to the appointing authority to consider initiating action against the officer who authored the October 2025 order.

Key Takeaway

The decision underlines and reiterates the responsible AI philosophy, ruling that AI-generated legal research cannot be relied upon in adjudication without independent verification.

The Court’s findings distinguish between the use of AI as an assistive tool and the substitution of AI-generated material for the decision-maker’s own legal analysis. Where an adjudicatory order relies upon non-existent authorities, fake citations, or legal propositions that the cited judgments do not actually establish, such reliance can undermine the sustainability of the order, along with justice itself.

The decision therefore places the responsibility for verification, legal reasoning and adjudication firmly on the human decision-maker, even where AI is used as an aid in the process.