Vijay Ghanshyam Gadiya v. Union of India & Anr., 2026 INSC 947 (2 September 2026)
What the case was about
This case began as a challenge to one of the largest customs penalties ever imposed on an individual, but it ended as a landmark warning about the dangers of unchecked artificial intelligence in legal research. The Supreme Court of India was asked whether an adjudicatory order can survive when the officer who passed it relied on case laws and legal articles that were entirely fabricated by an AI tool—judgments that never existed, citations that were fake, and legal propositions that were pure hallucination. The Court’s answer was emphatic: a decision poisoned by such material is a nullity and cannot stand, no matter how serious the underlying allegations.
The key facts
In October 2025, the Additional Commissioner of Customs, Surat, passed an Order-in-Original imposing a penalty of ₹425,27,99,100 on the appellant, Vijay Ghanshyam Gadiya, under Section 114 of the Customs Act, 1962 (para 2). The officer found that the appellant had deliberately mis-declared a consignment of natural diamonds as lab-grown diamonds to evade the higher customs duty applicable to natural stones (para 2).
The appellant challenged the staggering penalty before the Gujarat High Court, but his appeal was dismissed on 20 January 2026 (para 2). Undeterred, he approached the Supreme Court. Before the apex court, the appellant levelled an extraordinary charge: the second respondent had relied on several AI-generated judgments and articles in the original penalty order. According to the appellant, these included case laws that were completely non-existent, citations that were fabricated, and legal propositions that had no basis at all in actual jurisprudence.
The Supreme Court examined the claim rather than brushing it aside as a routine or frivolous objection. After individually verifying the citations, the Court confirmed that the second respondent had indeed relied upon case laws that were either non-existent or carried fake citations. In some instances, the case laws existed but did not support the legal ratio attributed to them; the attributed propositions appeared to be the product of an AI hallucination (para 4).
The questions before the Court
The appeal raised two linked questions of principle. First, is an adjudicatory order legally sustainable when the authority has relied upon AI-generated fake, non-existent, or hallucinated precedents without any independent verification? Second, should the impugned penalty order—and the Gujarat High Court’s confirmation of it—be set aside, with the matter remanded for a fresh decision completely untainted by such material?
What the Court decided and why
The Supreme Court allowed the appeal. It held that reliance by an adjudicating authority on AI-generated fake, non-existent, or hallucinated precedents without independent verification strikes at the very heart of the adjudicatory process. Delivering the judgment for a bench also comprising Justice Sheel Nagu, Justice Dipankar Datta declared that such a decision is “no decision in the eyes of the law,” regardless of whether the bogus material had a direct or indirect bearing on the final outcome (para 5). The Court ruled that even if an “iota of fake or hallucinated material enters the decision-making process,” the order must be set aside because it violates the integrity and sanctity of adjudication (para 5).
The Court drew upon its recent decision in Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., 2026 SCC OnLine SC 1258. It reiterated a stance of “zero-tolerance” for the production, citation, or use of unverified AI-generated precedents (para 5). It clarified that its censure is directed specifically at the presentation of fake or hallucinated material as if it were a genuine court precedent, and not at the legitimate use of artificial intelligence as a research aid in the ordinary course of legal work.
Turning to the facts before it, the Court found that the second respondent’s reliance on dubious, AI-generated material proved “fatal to the sustainability of the order imposing penalty” on the appellant (para 7). The Additional Solicitor General did not contest this conclusion.
Consequently, the Supreme Court set aside both the Gujarat High Court’s order dated 20 January 2026 and the original penalty order dated 8 October 2025 (para 8). The proceedings were revived and remanded for a fresh adjudication by another officer of the same rank, expressly excluding the officer who had authored the tainted order (para 8). The Court also left it to the wisdom of the appointing authority to consider initiating action against the author of the original order as may be warranted in accordance with law (para 11).
Why it matters
This ruling erects a formidable wall around the integrity of judicial and quasi-judicial decisions. By declaring that the presence of even a trace of unverified, AI-fabricated precedent renders a decision legally void, the Court has sent an unambiguous message to the Bar and the Bench at every level: AI chatbots and large language models are not infallible research assistants, and every citation must be independently checked against authoritative sources before it is pressed into service.
For litigants, the judgment offers a powerful and novel remedy. If an adverse order rests on phantom precedents, it is void ab initio and must be redone from scratch, irrespective of whether the fake law directly influenced the outcome. For the administration, the signal is equally stark—officers who draft orders without verifying the authenticity of their sources risk not only reversal on appeal, but also disciplinary or other institutional action that the appointing authority may deem fit.
At the same time, the Court was careful to distinguish between the misuse of AI and its rightful use. The judgment is not a blanket indictment of artificial intelligence in modern legal practice; it is a surgical rejection of fabricated authority. In a rapidly digitising legal world where technology is increasingly used to manage heavy caseloads, the decision reinforces an ancient principle for the digital age: the tools may change, but the duty of human verification remains absolute.