AI-Assisted Quasi‑Judicial Orders: Mandatory Human Verification of Citations and Independent Reasoning (Faiz Enterprise v. State Tax Officer)

Court: Gujarat High Court
Coram: Hon’ble Mr. Justice A.S. Supehia & Hon’ble Ms. Justice Vaibhavi D. Nanavati
Date: 20-08-2026
Case: FAIZ ENTERPRISE THROUGH PROPRIETOR MAHETAR MAHIR FARUKBHAI v. STATE TAX OFFICER, UNIT-67, SURAT & ANR.

1. Introduction

The Gujarat High Court, in a writ proceeding (Special Civil Application), confronted a growing administrative-law problem: quasi-judicial tax orders being drafted with the aid of Artificial Intelligence (AI) tools, without verification of the legal sources relied upon. The petitioner, Faiz Enterprise, challenged a chain of GST actions—starting from a show cause notice and culminating in cancellation of registration and appellate confirmation—primarily on the ground that the adjudicating authority cited case-law that was either non-existent or irrelevant, apparently sourced through AI.

The central issues before the Court were:

• Whether the impugned quasi-judicial order suffered from legal infirmity due to reliance on AI-generated/incorrect case citations and ratios.
• What institutional safeguards should govern the use of AI by adjudicating and quasi-judicial authorities.
• What consequential relief should follow for the taxpayer where the foundational proceedings are tainted.

2. Summary of the Judgment

The Court took cognizance that the State Tax Officer had relied “exclusively” on AI-generated case laws and that the cited authorities were alleged to be non-existent or mismatched in citation/ratio. On the next date, the concerned officer remained personally present, tendered an unconditional apology, and disclosed that he was a probationary officer who had used AI in drafting the order and citing non-existent/irrelevant judgments. The Department, in response, produced administrative instructions issued by the Additional Commissioner of State Tax (Admin) prescribing a verification-and-oversight framework for AI use in adjudication.

On merits and relief, the State fairly proposed to revisit the matter by resorting to Section 108 of the Goods and Services Tax Act, 2017 and/or by issuing a fresh show cause notice. The Court:

• Quashed and set aside the show cause notice dated 30/10/2025;
• Quashed and set aside the cancellation order dated 15/12/2025;
• Quashed and set aside the order rejecting revocation dated 10/03/2026 (issued on 23/03/2026);
• Quashed and set aside the appellate order dated 25/05/2026 and GST APL-04 dated 26/05/2026;
• Directed the authority to issue a fresh notice, consider the taxpayer’s defence, and pass a reasoned order in accordance with law;
• Directed that the Department’s AI-use instructions be scrupulously followed, and stated that violation would amount to contempt of the Court.

3. Analysis

3.1 Precedents Cited

A striking feature of the case is that the “precedents” surfaced not as binding guidance but as the problem itself. In its order dated 13/08/2026 (reproduced in the final judgment), the Court recorded the petitioner’s submission that the State Tax Officer relied on case laws that did not “even remotely deal” with the issue and, more seriously, that one of them was “not in existence.” The cases referred to in that context were:

• State of Gujarat v. Aarbee Structures Pvt. Ltd., 2024 (4) TMI 951 — asserted before the Court to be “not in existence.”
• M/s. Sri Vinayaga Agencies v. Assistant Commissioner, 2023 (5) TMI 106 — asserted to have a different citation and inapplicable ratio to the petitioner’s case.
• Jyoti Tar Products Pvt. Ltd. v. Deputy Commissioner, 2024 (7) TMI 512 — asserted to not match citations and to have an inapplicable ratio.

The Court did not undertake a conventional exercise of weighing these precedents on merits; instead, it treated their alleged non-existence/irrelevance as symptomatic of a deeper defect: outsourcing adjudicatory reasoning to unverified AI output. The “influence” of these citations, therefore, lay in exposing the vulnerability of quasi-judicial orders to “hallucinated” authorities and prompting a governance response.

3.2 Legal Reasoning

The Court’s reasoning has two interconnected strands:

(A) Institutional integrity of quasi-judicial decision-making

The Court’s prima facie view (13/08/2026) was that the impugned order appeared to be based “exclusively on the AI generated case laws.” This directly engages core public-law requirements applicable to quasi-judicial authorities:

• Application of mind: the decision-maker must demonstrate independent evaluation of facts and law; it cannot be substituted by automated text.
• Reasoned order: reasons must be traceable to correct law and relevant material; incorrect or imaginary citations defeat legal sustainability.
• Judicial scrutiny readiness: administrative adjudication must be able to “withstand” scrutiny; reliance on unverifiable sources makes the order inherently fragile.

(B) Norm-setting: controlled use of AI with verification and accountability

The Department’s instructions (issued after the Court’s intervention) were incorporated into the judgment and then elevated by a judicial direction requiring strict compliance. The instructions crystallize the operative rule for the tax department’s adjudication culture:

• AI is only an assistive tool; it cannot replace the officer’s own reasoning.
• Any AI-identified case law/statute must be independently verified from primary official sources before reliance.
• Citations must be accurate; relevance must be confirmed; and the current status (including whether overruled) must be checked.
• If AI paraphrases a ratio, the officer must locate the original paragraph and reproduce it verbatim before relying.
• Taxpayer-cited judgments must be considered; if rejected, reasons must be recorded.
• Ultimate responsibility remains with the issuing authority; failure may be misconduct and attract disciplinary action.

(C) Remedial approach in the individual case

On relief, the Court accepted the State’s fair stand to “revise the entire orders” and proceeded to set aside the entire chain (SCN, cancellation, revocation rejection, and appellate order), directing a fresh notice and a fresh, reasoned decision after considering the petitioner’s defence—while keeping “all rights and contentions” open. This reflects a corrective, process-restorative remedy: the taxpayer is put back to a lawful hearing stage, and the administration is required to decide again under proper standards.

3.3 Impact

The judgment’s likely significance extends beyond the individual GST dispute:

• Administrative AI governance becomes justiciable: the Court did not merely criticise AI misuse; it anchored compliance to departmental instructions and attached the consequence of contempt for violation.
• Higher drafting and citation discipline in tax adjudication: officers are effectively required to verify sources, reproduce relevant extracts, and demonstrate relevance—raising the minimum standard for “legal sustainability.”
• Template for other departments: the principles—verification, human oversight, accountability—are transferable to any quasi-judicial setting using AI-assisted research/drafting.
• Litigation behaviour: taxpayers will likely scrutinize citations more aggressively; departments may implement stronger training, audit trails, and supervisory checks.

4. Complex Concepts Simplified

• Show Cause Notice (SCN): a notice asking a person to explain why proposed action (here, GST consequences like cancellation) should not be taken.
• Cancellation of GST registration: termination of a taxpayer’s registration, affecting the ability to legally collect tax and claim input tax credit, and often disrupting business operations.
• Revocation application: a request to restore a cancelled registration (subject to statutory conditions and timelines).
• Quasi-judicial authority: an administrative officer who decides disputes affecting rights and liabilities and must follow fairness, reasons, and lawful procedure.
• AI-generated case laws / “hallucinations”: AI tools can produce plausible-sounding citations or summaries that are incorrect or non-existent unless independently verified.
• Reasoned order: a decision that explains facts, law, arguments, and the logic connecting them, enabling appeal and judicial review.
• Contempt of Court: disobedience of a court’s directions can invite proceedings to enforce authority of the Court; here, the Court warned that violating the AI-use instructions would attract contempt consequences.

5. Conclusion

Faiz Enterprise v. State Tax Officer establishes a practical and enforceable rule for modern tax adjudication: AI may assist, but adjudication must remain a human, accountable function grounded in verified legal sources and independent reasoning. By setting aside the tainted proceedings, directing a fresh lawful process, and mandating compliance with departmental AI instructions (backed by the warning of contempt), the Gujarat High Court positions “verification, relevance, and application of mind” as non-negotiable standards in an era of AI-assisted drafting—thereby shaping how quasi-judicial administration must adapt to technology without surrendering legality.