The Gujarat High Court censured a GST officer for depending on unverified AI-produced court citations, which proved fictitious. The errors impacted a matter involving Marhabba Overseas Private Limited. The court called for thorough verification and prudent use of AI in legal processes, granting interim relief to the petitioner and signaling a cautious stance on AI-assisted quasi-judicial decisions.
The Gujarat High Court pulled up a Goods and Services Tax (GST) officer for relying on incorrect and non-existent judicial precedents while passing an order, a development which comes against the backdrop of the Supreme Court cautioning against increasing reliance on artificial intelligence (AI) in legal proceedings.
In a strongly worded order, the high court flagged the risks of unchecked reliance on AI in quasi-judicial functioning and criticised a central GST officer for citing judgments that either did not exist or were irrelevant to the issues involved.
A division bench of justices AS Supehia and Pranav Trivedi was hearing a writ petition filed by Marhabba Overseas Private Limited challenging an order dated September 26, 2025, passed by the additional commissioner, Central GST and Central Excise.
The bench observed that guidelines may be required to regulate how quasi-judicial authorities rely on precedents. It said “we find that this is an appropriate case, wherein some directions are called for regulating/prescribing some parameters for quasi-judicial authorities while placing reliance on the judgements either of the High Courts or of the Supreme Court of India”.
During the hearing, senior advocate SN Soparkar highlighted what he termed a “very worrying trend” in the impugned order — reliance on judgments that were either incorrectly cited, non-existent or unconnected with the petitioner’s defence.
One of the objections raised concerned the alleged failure to supply relied upon documents (RUDs) along with the show-cause notice uploaded on the GST portal. The officer rejected the contention citing Union of India vs Coastal Container Transporters Association, 2019 SCC OnLine SC 1744. However, the petitioner argued that the citation was incorrect and the judgment dealt with service classification and maintainability of writ petitions, not non-supply of RUDs.
On the issue of non-issuance of DRC-01A under Rule 142(A)of the CGST Rules, the officer relied on NKAS Service Pvt. Ltd. vs Union of India (2021-VIL-37-MAD), purportedly of the Madras High Court. The petitioner contended that no such Madras High Court ruling existed and that the relevant decision was in fact delivered by the Jharkhand High Court in favour of the assessee, holding that deficiencies in DRC-01A would violate principles of natural justice.
Similarly, reliance was placed on CC vs Flock India Pvt. Ltd., 2000 (120) ELT 285 (SC), which, according to the petitioner, related to classification and refund under the Central Excise Act and had no relevance to the case.
In another instance, the officer cited Rajasthan State Chemical Works vs Union of India, 1991 (55) ELT 444 (Guj), which the petitioner argued did not exist as a Gujarat High Court judgment. The reported decision, counsel submitted, was in fact a Supreme Court ruling and unrelated to the issue of natural justice raised in the case.
Taking serious note, the bench described the findings as “flawed and deceptive” and observed that it appeared the officer had followed AI-generated citations without reading the actual judgments.
“The reasonings/findings recorded by the respondent-Commissioner… is flawed and deceptive. It appears that the Commissioner, without reading the actual judgements, has followed the AI generated citations and case law,” the court observed.
The high court granted interim relief to the petitioner till final disposal of the writ petition.
Commenting on the order, Ikesh Nagpal, lead–indirect tax at AKM Global, said the ruling should serve as a wake-up call. “Though artificial intelligence is a powerful research assistant, it is not an adjudicator. The moment a quasi-judicial authority substitutes independent judicial application of mind with AI-suggested citations, the very architecture of natural justice begins to erode,” he said, adding that the court intervened promptly to preserve the integrity of the adjudicatory process.”
Please click here to view the full story on Economic Times.