The Supreme Court of India canceled a $45 million fine on September 2, 2026, after discovering that a customs officer had cited fabricated court rulings and fake quotes generated by artificial intelligence. The decision extends a "zero tolerance" policy the court established just two months earlier, making clear that any legal ruling containing AI hallucinations will be annulled regardless of how central the fake material was to the outcome.
The Additional Commissioner of Customs imposed the fine on Vijay Ghanshyam Gadiya after determining that a batch of natural diamonds had been declared as lab-grown diamonds to apply a lower tariff. Gadiya appealed the fine in the Gujarat High Court, which rejected his appeal in January 2026. He then turned to the Supreme Court.
A bench consisting of Justices Dipankar Datta and Sheela Nagu did not address the substantive issue of possible misdeclaration of goods. Instead, the court identified three types of errors in the customs officer's decision: some cited cases did not exist at all, some contained fake references, and real court rulings were used to justify legal conclusions they did not actually contain.
Three categories of fabricated legal material
The last category of errors drew particular attention from the bench. The court described cases where genuine rulings were twisted to support arguments they never made as an obvious "AI hallucination." This distinction matters because it shows the court is developing a taxonomy for AI-generated errors in legal documents, separating complete fabrications from distorted real rulings.
The Supreme Court annulled both the customs authority's decision and the High Court's ruling. The case was remanded for reconsideration by another official of the same rank. The appointing authority also received the right to decide on possible measures regarding the customs officer who prepared the defective decision.
The precedent set in July 2026
This was not the first time India's highest court confronted AI-generated fake legal material. In July 2026, the Supreme Court dealt with a similar issue in Pooja Ramesh Singh v Jammu and Kashmir Bank Ltd. That case involved six allegedly judicial precedents used by the National Company Law Tribunal. Among them were completely fabricated rulings, incorrect references, and non-existent paragraphs attributed to genuine court decisions.
At that time, the Supreme Court applied a "zero tolerance" principle to AI-generated fake court rulings. The court said a decision containing even a small amount of fabricated or "hallucinated" legal material must be annulled regardless of whether this material directly affected the outcome of the case. The September ruling applies that same standard to customs proceedings.
The court's view on AI in legal work
The Supreme Court acknowledged that artificial intelligence can assist in legal research and improve work efficiency. But it drew a hard line: AI tools are acceptable only when a human controls the decision-making process and verifies the results obtained. This position aligns with broader discussions in the legal profession about AI for Legal applications and the boundaries of responsible use.
For paralegals and legal researchers who increasingly encounter AI-powered research tools, the court's standard creates a clear professional obligation. Verification is not a best practice - it is a requirement that carries serious consequences when ignored. The AI Learning Path for Paralegals addresses these verification protocols alongside document review and legal research automation techniques.
Why this matters for legal professionals
The Supreme Court of India has now twice ruled that AI hallucinations in legal documents are fatal to the decision containing them. There is no safe harbor for arguing the fake material was incidental. For lawyers, judges, and customs officials, the message is blunt: cite a hallucinated case and your entire ruling collapses. The practical implication is that every legal professional who uses AI research tools must build verification into their workflow - checking every citation, every quote, and every paragraph attribution against primary sources before filing. The court will not distinguish between a careless human error and an unchecked AI output. Both produce the same result: an annulled decision and potential professional consequences for the person who submitted it.
Your membership also unlocks:
