Raj Jaggi
AI Enters GST Proceedings - A New Challenge for Tax Administration
Artificial Intelligence is increasingly becoming part of professional decision-making. Legal research, drafting, summarisation and identification of judicial authorities can now be undertaken with extraordinary speed. Tax administration cannot remain untouched by this technological change. Used responsibly, AI can assist tax authorities in locating statutory provisions, identifying relevant precedents, organising complex records and preparing drafts with greater efficiency.
The concern begins when technological assistance enters the statutory decision-making process without adequate verification and human scrutiny. A show cause notice is not merely a compilation of allegations, facts and legal provisions; it reflects the formation of a statutory view by the proper officer and requires application of mind to the material before him. An adjudication order carries an even greater responsibility, as it determines rights and liabilities after considering the taxpayer's defence. At both stages, the authority empowered by law must understand the facts, verify the legal material relied upon and independently apply the law.
This emerging concern recently came before the Gujarat High Court in Faiz Enterprise through Proprietor Mahetar Mahir Farukbhai v. State Tax Officer, Unit-67, Surat & Another, 2026-VIL-932-GUJ, R/Special Civil Application No. 10123 of 2026, decided on 20.08.2026. The controversy arose because the impugned GST proceedings relied upon AI-generated judicial authorities that were either non-existent or irrelevant. The High Court found the resulting decision-making process fundamentally defective and procedurally unsustainable.
The importance of the issue, however, extends beyond fabricated citations or inaccurate legal research. Read alongside the emerging judicial approach to AI-assisted notices, it raises a broader question for GST administration: how far can Artificial Intelligence participate in a statutory process that requires the designated authority itself to examine the material, form the necessary satisfaction and ultimately take responsibility for the decision? That question lies at the heart of the developing relationship between Artificial Intelligence and lawful tax administration.
The Controversy - When AI-Generated Authorities Entered the Order
The dispute arose from proceedings culminating in the cancellation of the petitioner's GST registration by an order dated 15.12.2025. What initially appeared to be a routine challenge to cancellation assumed much wider significance when the judicial authorities relied upon in the order were scrutinised before the Gujarat High Court.
The petitioner pointed out that one of the decisions cited by the State Tax Officer apparently did not exist, while other authorities had been cited with incorrect particulars or for propositions that did not emerge from those judgments. In particular, State of Gujarat v. Aarbee Structures Pvt. Ltd., cited as 2024 (4) TMI 951, was stated to be non-existent. Similar objections were raised regarding M/s Sri Vinayaga Agencies v. Assistant Commissioner and Jyoti Tar Products Pvt. Ltd. v. Deputy Commissioner, on the ground that the citations or the legal principles attributed to those decisions did not correspond with the authorities actually relied upon.
The problem, therefore, went beyond an occasional error in citation. The very legal foundation on which the adjudicating authority had proceeded became doubtful. Judicial precedents had apparently been reproduced without first verifying whether they existed, whether their citations were correct, and, more importantly, whether their ratio had any application to the controversy being decided.
Taking note of this serious issue, the High Court, by its order dated 13.08.2026, recorded a prima facie impression that the impugned order appeared to have been passed by placing exclusive reliance on AI-generated case law and called upon the officer to obtain instructions. What began as a dispute concerning the cancellation of GST registration thus developed into a larger question of considerable importance: can a statutory adjudication survive when the legal authorities forming its foundation are generated by AI but remain unverified by the officer who relies upon them?
Beyond the Officer's Apology - The Larger Risk of Unverified AI
At the subsequent hearing, the officer who had passed the impugned order filed an affidavit tendering an unconditional apology. He acknowledged using Artificial Intelligence while drafting the order and accepted that some of the judicial authorities cited therein had subsequently been found to be non-existent or irrelevant. He explained that he was a probationary officer and that the lapse had occurred due to lack of experience.
The affidavit disclosed another significant development. Following the High Court's earlier order, the officer had attended a training programme on the use of Artificial Intelligence in drafting special civil applications as well as adjudication and appellate orders. This indicated that the issue was no longer confined to an isolated mistake. As AI increasingly becomes part of legal and administrative work, the more important question is how it should be used without compromising the safeguards inherent in statutory adjudication.
Generative AI can produce legal language with remarkable fluency. It can suggest case law, summarise judgments and formulate propositions that closely resemble conventional legal research. But the apparent confidence of an AI-generated response is no assurance of its legal accuracy. A citation may be fabricated; an existing judgment may be attributed a proposition it never laid down; or an otherwise relevant precedent may have been distinguished, overruled or rendered inapplicable by subsequent developments.
This risk assumes particular importance in tax adjudication. An order may appear polished, detailed and legally reasoned, yet remain fundamentally defective if the authorities on which it rests have not been independently verified. The real concern, therefore, is not the use of AI itself, but the substitution of AI-generated material for the adjudicating officer's own verification, reasoning and independent application of mind.
AI May Locate the Law - Verification Must Remain Human
The High Court's intervention was followed by an immediate administrative response. On 18.08.2026, the Additional Commissioner of State Tax issued detailed instructions regulating the use of Artificial Intelligence by adjudicating and quasi-judicial authorities. The instructions recognise that AI may be useful as a research and drafting aid, but its output cannot be accepted at face value.
The central requirement is independent verification. Any material identified or generated through AI—whether a judicial precedent, statutory provision, rule, circular or notification—must be verified against an authentic source before it forms part of an adjudicatory decision. Where a judgment of a superior court is proposed to be relied upon, the authority must verify not merely its citation but also its relevance to the controversy and its continuing legal validity.
Verification of case law necessarily goes beyond confirming that a judgment exists. The adjudicating authority must examine the decision itself, understand the factual and legal context in which it was rendered, and ascertain whether the proposition attributed to it actually forms part of its ratio. It must also be ensured that the decision has not subsequently been overruled, reversed, or otherwise deprived of its precedential value.
The distinction is fundamental. AI may help an officer locate a judgment, but it cannot discharge the officer's responsibility to verify, understand, and correctly apply that judgment. Technology can accelerate legal research; the responsibility for determining whether the law so discovered is authentic, relevant, and applicable must remain with the statutory authority. The technology can accelerate the search, but verification remains a human responsibility.
An AI Summary Cannot Replace the Judgment Itself
Another important safeguard in the Gujarat SGST instructions dated 18.08.2026 concerns AI-generated summaries of judicial decisions. AI tools can quickly condense lengthy judgments and identify what appears to be the principle emerging from them. Such assistance may be useful at the preliminary stage of legal research, but an adjudicatory order cannot safely rest on a machine-generated summary alone.
The instructions therefore require that, where an AI tool paraphrases the ratio of a judgment, the officer must locate and examine the relevant paragraph of the original decision before relying on it. This requirement is important because the ratio of a judgment cannot always be understood from an isolated sentence or a condensed summary. Judicial observations derive their meaning from the facts, the statutory provisions under consideration and the precise controversy before the court.
This assumes particular significance under GST, where seemingly small distinctions may materially affect the applicability of a precedent. The tax period may differ; the statutory provision may have been amended; or the earlier judgment may concern registration while the pending dispute relates to input tax credit, valuation or some entirely different issue. Even where the same statutory provision is involved, a precedent cannot automatically be applied unless the underlying factual and legal context is comparable.
Case Law Cited by the Taxpayer Must Also Be Addressed
The Gujarat SGST instructions dated 18.08.2026 extend beyond verifying authorities identified through Artificial Intelligence. They specifically require adjudicating authorities to consider and discuss the judicial decisions relied upon by the taxpayer. If any such decision is found distinguishable or inapplicable, the order must record the reasons for that conclusion.
This requirement reflects a fundamental principle of reasoned adjudication. Independent application of mind is not demonstrated merely by citing authorities that support the Department's proposed view. The submissions and precedents relied upon by the taxpayer form part of the defence and must receive meaningful consideration. A judicial authority cannot be brushed aside merely because it does not support the conclusion the adjudicating officer is otherwise inclined to reach.
Human Oversight Must Remain at the Heart of AI-Assisted Adjudication
The Gujarat SGST instructions dated 18.08.2026 do not discourage the use of Artificial Intelligence. On the contrary, they recognise that AI tools may improve the efficiency and effectiveness of adjudication. The regulatory approach is therefore not to exclude technology from tax administration, but to ensure that its use remains consistent with the requirements of lawful decision-making.
The essential safeguard is meaningful human oversight. Recognising concerns about the accuracy, transparency and underlying information used by AI tools, the instructions require appropriate human supervision of AI-generated material. More importantly, every adjudication order must reflect the officer's own application of mind and independent legal reasoning based on the facts and law governing the dispute. AI may assist that process, but it cannot perform the statutory function entrusted to the adjudicating authority.
This is perhaps the most important principle emerging from Faiz Enterprise. The relevant question is not whether an officer has used AI in researching or drafting an order. Technology is increasingly part of modern administration, and its responsible use can undoubtedly improve efficiency. The real test is whether the statutory authority has independently examined the facts, verified the law, considered the taxpayer's defence and arrived at its own reasoned conclusion.
The Authority Signing the Order Remains Accountable
The Gujarat SGST instructions dated 18.08.2026 make another important principle explicit: the ultimate responsibility for the correctness of an adjudication order rests with the authority issuing it, irrespective of the technological tools used in its preparation. Failure to observe the prescribed safeguards, including reliance on unverified AI-generated material, may amount to misconduct and invite disciplinary action.
This principle extends beyond the use of Artificial Intelligence. A statutory authority cannot distance itself from an erroneous order by pointing to the source of the error whether a subordinate prepared the draft, an electronic database supplied an incorrect citation, or an AI tool generated inaccurate legal reasoning. Once the order is issued in the authority's name and designation, responsibility for its contents remains with that authority.
When Administrative Safeguards Acquire Judicial Force
The significance of Faiz Enterprise extends beyond the issuance of Gujarat SGST instructions dated 18.08.2026. The Gujarat High Court did not merely take note of those instructions; it directed that they be followed scrupulously and made it clear that their violation would amount to contempt of the Court. The safeguards governing the use of Artificial Intelligence thus acquired a much stronger practical footing in GST adjudication within the State.
The requirements themselves are neither complex nor hostile to technology. They essentially demand what sound adjudication has always required: verification from authentic sources, accurate citation of authorities, examination of their relevance and continuing legal validity, proper consideration of precedents relied upon by the taxpayer, meaningful human oversight and independent application of mind by the adjudicating authority.
Defective Adjudication Vitiated the Entire Proceedings
The consequences of the defect at Faiz Enterprise were not confined to disregarding an incorrect citation or correcting the cancellation order. The Gujarat High Court found the flaw sufficiently fundamental to set aside the entire chain of proceedings that followed the defective adjudicatory process.
Accordingly, the show cause notice dated 30.10.2025, the cancellation order dated 15.12.2025, the order rejecting the application for revocation dated 10.03.2026 and issued on 23.03.2026, the appellate order dated 25.05.2026, and GST APL-04 dated 26.05.2026 were all quashed. This restored the matter to the stage from which lawful proceedings could commence afresh.
The authority was permitted to issue a fresh notice, while the taxpayer was required to cooperate with the proceedings. Its reply and defence were to be duly considered before passing a fresh reasoned order in accordance with law, with all rights and contentions left open. The Court thus corrected the defective process without preventing the Department from taking lawful action on the underlying matter.
The nature of the relief is particularly instructive. Improper use of AI did not confer immunity upon the taxpayer from proceedings otherwise permissible under law; it rendered unsustainable the process through which the adverse decision had been reached. The defect was cured not by insulating the taxpayer from adjudication, but by requiring the adjudication to begin afresh and proceed in the manner contemplated by law.
The Same Judicial Concern Had Emerged Earlier
The principle emerging from Faiz Enterprise finds support in an earlier decision of the Punjab & Haryana High Court in M/s SRO India v. State of Punjab and Another, 2026-VIL-724-P&H, Neutral Citation 2026:PHHC:091427-DB, CWP-11494-2026 (O&M), decided on 07.07.2026. Though arising in a different factual setting, the decision had already highlighted the risks of allowing AI-assisted material to replace the statutory authority's own decision-making process.
In SRO India, Form GST DRC-01A and the accompanying material uploaded on the portal revealed traces of AI-assisted drafting. The material contained suggestions for improving the proposed reasoning and preparing the order. The taxpayer questioned the proceedings on the ground that the statutory authority had not independently examined the matter before initiating action.
The Punjab & Haryana High Court emphasised that the statutory scheme requires the competent authority to examine the facts and apply its mind before issuing a notice. Finding that the impugned notice had primarily relied upon an AI tool, the High Court quashed Form GST DRC-01A and the consequential proceedings. At the same time, it preserved the liberty of the competent authority to proceed afresh after independently examining the facts and applying its own mind.
The two decisions address different manifestations of the same underlying concern. SRO India dealt principally with AI-assisted preparation of a notice where independent application of mind was found wanting. Faiz Enterprise exposed an additional danger— AI-generated judicial authorities entering an adjudication order without proper verification. Read together, the two judgments draw an emerging judicial line: AI may assist a statutory authority in research and drafting, but it cannot replace the human judgment, verification and independent application of mind that the law entrusts to that authority.
Independent Application of Mind - A Principle Beyond AI
The significance of Faiz Enterprise and SRO India should not be confined to cases involving Artificial Intelligence. The requirement of independent application of mind is not a new safeguard created in response to AI; it is a long-established requirement of lawful quasi-judicial decision-making. AI has merely introduced a new setting in which that principle assumes greater practical importance.
The same legal concern can arise even without technology. An adjudicating authority may mechanically reproduce the conclusions of an investigation report, copy reasoning from another order, adopt a draft prepared by a subordinate or proceed without genuinely considering the taxpayer's explanation. In each situation, the defect is essentially the same: the person upon whom the statute confers the power to decide has not independently exercised that power.
The source of assistance is therefore not decisive. Research, drafting and organisational support may come from officers, databases, precedents or technological tools. What ultimately matters is whether the statutory authority has itself examined the relevant facts, considered the defence, verified the applicable law and reached its own reasoned conclusion.
Author's Perspective - AI May Assist Adjudication, but the Decision Must Remain Human
Faiz Enterprise and SRO India together provide an important framework for the responsible use of Artificial Intelligence in tax adjudication. Neither decision rejects technology. What they insist upon is that AI-generated material must be independently verified and that the statutory authority must itself examine the facts, ascertain the applicable law, consider the taxpayer's defence, and arrive at its own reasoned conclusion.
The principle may therefore be stated simply: technology may assist the decision-making mind, but it cannot become the decision-maker. Ultimately, the validity of an adjudicatory order must depend not on what AI suggested, but on what the proper officer independently examined, verified, and decided—and on the reasons for that decision.
CAclubindia
