When the Notice is AI-Assisted but the Officer isn't: SRO India and the Case against Mechanical Adjudication in GST

 

Ashwarya Sharma, Advocate | Co-Founder & Legal Head, RB LawCorp

 

We all know that show cause notices have long been drafted from templates, borrowed language, and boilerplate reasoning none of that is new. What is new is a tax officer turning to an AI tool to do the drafting itself, and a taxpayer discovering, on the department's own portal, a record of that exchange. In SRO India v. State of Punjab and Another, 2026-VIL-724-P&H, the Hon'ble Punjab & Haryana High Court was presented with exactly this situation. The notice under challenge had, appended to it, the AI tool's own suggestions for strengthening the department's case, including an offer to make the eventual order "lethal".

 

This time the petitioner did not need to argue that the officer had failed to apply his mind in the abstract as the evidence of how the notice had actually been produced was sitting in the record. The High Court's response was brief but unambiguous and it settles a question indirect tax practitioners have been circling for some time that a notice that carries a machine's suggestions instead of an officer's own reasoning is not a notice the law recognises at all. As AI drafting tools become routine in tax offices, this judgment is likely to be the first of several tests of where assistance ends and abdication begins.

 

1. Facts Leading to the Dispute

The petitioner, was served a pre-SCN notice in Form GST DRC-01A which is the precursor to formal proceedings under the CGST Act. On the face of it, the notice looked like any other. What set this case apart was a document annexed to the notice and separately uploaded on the GST portal, which did not read like the work of a tax officer at all. It referred to "various AI-based aids" and carried suggestions on how the order arising from the notice could be improved or made more forceful. The Court reproduced the passage in its judgment, and it is worth setting out in full, because it tells the story better than any submission could:

 

"Optional Enhancement (for Order-in-Original) - If you want, I can: Tighten this further with case-law citation paragraph-wise; Convert it into Order-in-Original reasoning; Add 'knowledge and connivance inferred from facts' language; Draft a defence-proof rebuttal against 'buyer not responsible' plea. Just say 'add OIO version' or 'make it lethal'. Enhanced Verification Findings on Non-Movement of Goods (Vehicle-wise Analysis)."

 

2. The Petitioner's Case

It was obviously argued by the petitioner that there was no application of mind to the facts of the case by the officer concerned, who had instead resorted to an artificial intelligence tool for issuing the notice. The suggestions embedded in the notice was enough to demonstrate that the substance of the notice had been shaped by the AI-software rather than by the statutory authority which the law requires to form an independent opinion.

 

3. The State's Explanation

The State argued that reference to AI at the tail end of the notice was an inadvertent error rather than evidence that the notice itself was AI-authored.

 

4. The High Court's Analysis

 

4.1 Application of Mind Cannot Be Outsourced

The Bench was unimpressed by the State's explanation. It held that the statute expects the competent authority not merely to examine the facts of the case, but to apply his or her own mind before issuing a show cause notice. An inadvertent upload, even if true, does not answer the more basic question of how AI-generated material came to exist in relation to the notice in the first place.

 

4.2 AI-Drafted Notices Have No Statutory Sanction

The Court importantly held that the use of an AI tool for the preparation and issuance of a show cause notice had not been shown to be permissible under the statute and reliance on an AI tool for this purpose lacks the requisite sanction in law.

 

4.3 Quashing the Notice and All Consequential Proceedings

Having found that the show cause notice had been issued primarily relying upon the AI tool, the Court held that it could not be sustained in law and was rightly quashed.

 

5. The Legislative Scheme Behind the Verdict: Grounds, Speaking Notices, and Application of Mind

Tax professionals will remember, that this judgement did not emerge from a vacuum but sits on a foundation that indirect tax jurisprudence has built over decades around the show cause notice as the department's starting point and outer limit. The Hon'ble Supreme Court way back in Commissioner of Central Excise v. Brindavan Beverages (P) Ltd., 2007-VIL-45-SC-CE, has famously held that "show cause notice is the foundation on which the department has to build up its case. If the allegations in the show cause notice are not specific and are on the contrary vague, lack details and/or unintelligible that is sufficient to hold that the Noticee was not given proper opportunity to meet the allegations indicated in the show cause notice." This principle has been reiterated in a catena of decisions under both the GST law and the erstwhile indirect tax enactments, all converging on the same point that a notice must contain sufficient particulars of the allegations and the basis of the proposed demand to allow the noticee an effective and meaningful reply. A vague, cryptic, or non-speaking notice defeats this purpose and falls foul of natural justice.

 

The CGST Act itself legislates this principle rather than leaving it to inference. The newly drafted Section 74A(1) which talks about raising of demand under the CGST Act post FY 2024-25 onwards clearly requires the proper officer to serve notice of demand in which sub-sections (3) and (4) permit the proper officer, in lieu of a fresh and complete notice for a different tax period, to serve merely a statement containing the gist of the grounds already set out in an earlier notice only where the grounds relied upon remain the same. Section 75(7) closes this loop by stating that amount demanded in the final order cannot exceed what was specified in the original notice, and no demand can be confirmed on grounds other than those specified in the notice. Read together, this scheme makes 'grounds' the essence of a valid show cause notice; the statute permits statement only where the grounds are identical, which is itself proof of how central the grounds are to the document's validity. It follows that the reasoning behind a demand must be discernible from the notice as it stands, not reconstructed from material lying outside it.

 

Therefore, a proper officer cannot satisfy Section 74A and 75 by reproducing an AI template without independently examining and verifying the facts of the particular taxpayer's case. A notice or order built this way is not, in substance, an act of the statutory authority at all and is arbitrary and void ab initio.

 

6. The Deeper Malaise: Mechanical Confirmation, Tribunal Backlog, and What AI Threatens to Accelerate

The problem SRO India exposes is not new and the case has simply made visible what has been happening for years without leaving a trail. The GST Appellate Tribunal is reported to have commenced its functioning with a backlog upward of four lakh pending appeals. It may be recalled that at its predecessor CESTAT (Customs, Excise and Service Tax Appellate Tribunal), by most accounts, the department succeeds in barely 1 or 2 out of every ten matters it contests, with the taxpayer prevailing in the overwhelming majority. These are not isolated statistics as they point to a pattern at the adjudication and first appellate stage by departmental officers. Too often, the reply filed by a taxpayer is not meaningfully engaged with, and the demand proposed in the notice is confirmed as a matter of course, regardless of what the reply actually says. This runs against the basic expectation of a quasi-judicial authority, which is meant to weigh the material before it and decide without bias or a predetermined outcome, not to treat confirmation of the demand as the default. It is this pattern of mechanical confirmation, repeated across thousands of orders, that has fed the pendency now sitting at the GSTAT and pushed an increasing share of these disputes up to the Hon'ble High Courts and the Supreme Court.

 

Seen against this backdrop, the entry of AI into notice and order drafting is less a new risk than an accelerant on an old one. If the underlying habit is to confirm the demand without genuinely engaging with the taxpayer's case, a tool that can generate confident-sounding reasoning in seconds does not fix that habit, it multiplies it. Rather than easing the load on an already overstretched appellate system, unchecked use of AI in adjudication threatens to multiply the very orders that end up being challenged and set aside, adding further weight to courts and tribunals that are struggling with the backlog as it stands.

 

7. Concluding Observations

Before parting it may be remembered that Courts have long insisted that a show cause notice reflect an officer's own reasoning, not a rubber stamp, and SRO India confirms that this requirement does not bend for AI any more than it bent for templates or copy-pasted departmental language before it.

 

The case also points to an upcoming pandemic so to say. A Tribunal starting out with a backlog running into lakhs of appeals, and a department that loses the bulk of what it contests at Tribunal level, trace back to the same root cause that demands are confirmed without genuinely engaging with the taxpayer's reply expected from an independent, unbiased quasi-judicial authority. AI does not create this habit, but left unchecked, it risks making it faster and more routine, adding to a pendency problem the system can ill afford.

 

[Date: 22/07/2026]

 

(The views expressed in this article are strictly personal.)