Tax officers in Gujarat could face contempt proceedings if they use unverified material generated by artificial intelligence in adjudication or appellate orders, the Gujarat High Court has warned after a GST dispute exposed the use of fabricated and irrelevant judicial precedents in an official decision. The warning came in a case involving Faiz Enterprise, whose GST registration had been cancelled following proceedings before the State tax authorities. While challenging the action, the firm drew the High Court’s attention to judgments cited by the tax officer and questioned whether the authorities relied upon actually supported the conclusions reached against it.
Scrutiny of the citations revealed serious problems. One judgment referred to in the tax order could not be found at all, while other decisions were cited inaccurately or dealt with issues unrelated to the dispute. A bench of Justices AS Supehia and Vaibhavi D Nanavati had on August 13 taken note of the discrepancies and sought an explanation from the State. The court indicated that action could follow if it emerged that the officer had based his decision on authorities generated by AI without checking whether they were genuine or relevant.
Officer admits AI use
When the case returned to court on August 20, State Tax Officer Devang Arvindkumar Yadav appeared before the bench and acknowledged using AI while preparing the order. Yadav, who described himself as a probationary officer, apologised unconditionally and attributed the error to inexperience. He also informed the court that he had subsequently undergone training dealing with the use of AI in drafting and tax adjudication.
By then, the State tax administration had responded to the episode by issuing instructions applicable across Gujarat. The August 18 directive sets out safeguards for officers who use AI or AI-assisted research while dealing with tax cases. The High Court endorsed those safeguards and ordered strict compliance. It went a step further by saying that breach of the instructions, now incorporated into its judicial directions, could invite contempt proceedings.
AI cannot replace legal reasoning
Under the new framework, an officer cannot simply reproduce a judgment, statutory provision, circular, rule or notification suggested by an AI system. Every such reference must first be checked against an authoritative primary source. Officials must also establish that a judgment is relevant to the dispute before them and verify whether it remains good law. Where an AI tool provides a paraphrased version of a judicial ruling, the officer is required to go back to the original decision before relying on the proposition.
The safeguards also address material submitted by taxpayers. Judicial precedents cited by an assessee must be considered, and an officer who decides that a ruling does not apply must explain why.
Most importantly, responsibility for an order remains with the official who signs it. Technology can assist research, but the final decision must demonstrate the officer’s independent assessment of the facts and applicable law. The State’s instructions provide that reliance on unchecked AI material can also constitute misconduct and lead to disciplinary proceedings.
Fresh proceedings ordered
The controversy also resulted in the tax action against Faiz Enterprise being effectively reset.
The State told the High Court that it was prepared to revisit the disputed proceedings and issue a fresh decision. The bench consequently set aside the show-cause notice issued in October 2025, the subsequent cancellation of the firm's GST registration, the rejection of its attempt to have the registration restored and the appellate decision that followed.
Tax authorities have been permitted to begin the process afresh. They must consider the firm's defence and pass a reasoned order in accordance with law.
The episode highlights a growing problem accompanying the adoption of generative AI in legal work: systems can produce citations that appear credible but refer to judgments that do not exist or do not support the proposition for which they are cited.
The Gujarat High Court has separately adopted an AI policy that prohibits unverified AI-generated case references and statutory material and restricts the technology's role in substantive judicial decision-making.
The latest ruling effectively extends the emphasis on verification and human accountability to tax adjudication, making clear that technological assistance cannot dilute an officer’s responsibility for the legal accuracy of an official order.