Rs. 27.91 Crore Tax Demand Quashed: The Assessing Officer Who Used AI to Find Fake Judgments

KMG Wires Private Limited v. Income Tax Officer — Bombay High Court, October 2025

Hallucinations & Sanctions | India — Bombay HC | Income Tax

① CASE AT A GLANCE

Full Citation	KMG Wires Private Limited v. ITO & Ors., WP(L) No. 24366/2025, Bombay HC
Court	Bombay High Court (Writ — Larger Bench)
Judges	Justice G.S. Colabawalla and Justice Kamal Khata
Date of Decision	October 2025
Category	AI Hallucination / Income Tax Assessment
AI Tool Used	Not disclosed (AI research tool used by AO)
Who Used AI	Income Tax Assessing Officer (quasi-judicial authority)
Hallucinations Found	3 completely non-existent judgments cited in the assessment order
Assessment Demand	Rs. 27,91,00,000 (Rs. 27.91 Crores)
Outcome	Assessment ORDER QUASHED AND SET ASIDE; matter remitted for fresh consideration
	https://www.damiencharlotin.com/documents/925/

② BACKGROUND

KMG Wires Private Limited, a Maharashtra-based company, faced an income tax assessment for a particular assessment year in which the Assessing Officer (AO) raised an addition of Rs. 27.91 Crores. The AO’s order was supported — purportedly — by judicial precedents: three case citations that were incorporated into the assessment order to justify the legal position taken.

When KMG Wires challenged the assessment before the Bombay High Court by way of a writ petition under Article 226 of the Constitution of India, the petitioner’s counsel raised a preliminary challenge: each of the three cited judgments could not be traced in any official legal database, court record, or authoritative reporter. The citations appeared to have been fabricated.

Investigation of the source revealed that the AO had used an AI tool for legal research and had included the AI-generated citations in the official assessment order without verification. The three judgments did not exist anywhere in the Indian legal system.

③ THE AI ISSUE

The legal issue before the Bombay HC was whether an income tax assessment order — a quasi-judicial order with serious financial consequences — can be sustained when the legal reasoning in the order is premised entirely on judgments that do not exist.

The secondary issue: Does the Income Tax Act, 1961 or the principles of natural justice and administrative law require that the authority cited in quasi-judicial orders be real and verifiable? And what is the consequence for the taxpayer — and for an the AO — when fabricated citations are used to create a multi-crore tax demand?

④ WHAT THE COURT DECIDED

• The Bombay HC held that an assessment order citing non-existent judgments as legal authority has no legal foundation — it is an order without reasons within the meaning of Section 250 read with the principles of natural justice [audi alteram partem / right to reasoned orders].
• The court rejected any suggestion that the fabrication was a harmless typographical or administrative error. Citing non-existent judgments in a formal quasi-judicial order constitutes a fundamental illegality going to the validity of the order, not a mere irregularity.
• The entire assessment order of Rs. 27.91 Crores was QUASHED AND SET ASIDE — the company was freed from the demand, and the matter was remitted to the AO for fresh assessment after complying with the principles of natural justice.
• The HC signalled that AOs who use AI tools to generate legal citations without verification would face the consequence of having their orders invalidated — and potentially face departmental action.
• The judgment adds to a growing line of Indian authority — alongside the Gummadi and Jeetmal Choraria cases — establishing that AI hallucinations in quasi-judicial and judicial proceedings will be treated as fundamental illegalities, not correctable errors.

⑤ KEY QUOTE FROM THE JUDGMENT

“The citations relied upon by the Assessing Officer do not exist in law. An authority that has no existence cannot be a basis for a quasi-judicial order affecting a citizen’s rights. The order is accordingly quashed.”

— Justice G.S. Colabawalla, Bombay High Court, WP(L) 24366/2025, October 2025

⑥ THE INDIA ANGLE

Indian Law Equivalent

The KMG Wires case sits at the intersection of income tax law, constitutional law, and principles of natural justice:

• Section 143(3) of the Income Tax Act, 1961 — The AO’s assessment order under Section 143(3) must be based on reasons that are legally sustainable. Citing non-existent precedents renders the ‘reasons’ a nullity.
• Article 226 of the Constitution of India — A writ of certiorari lies to quash a quasi-judicial order that is based on an error apparent on the face of the record. Citing fabricated judgments is precisely such an error.
• Principles of Natural Justice (Audi Alteram Partem) — For a taxpayer to respond meaningfully to an assessment, the legal basis must be real and accessible. Fake citations deny the taxpayer the right to understand and challenge the legal basis of the demand.
• Section 263 of the Income Tax Act — The Commissioner (PCIT) has power to revise orders that are erroneous and prejudicial to the interest of revenue. An AO using AI hallucinations could equally expose their orders to revision under this provision.

Bar Council of India Rules

While Bar Council Rules apply to advocates and not directly to AOs, this case has implications for tax practitioners. A chartered accountant or advocate representing a taxpayer who knowingly relies on AI-generated citations in submissions before the Income Tax Department or before the ITAT could face professional consequences analogous to those faced by advocates before courts.

The Institute of Chartered Accountants of India (ICAI) has not yet issued specific guidance on AI use in tax representations. The KMG Wires judgment makes a strong argument for ICAI to do so — particularly for AOs and tax practitioners who use AI tools in drafting assessment orders, submissions, and representations.

Practical Advice for Indian Advocates

• When challenging an income tax assessment, always check the citations in the AO’s order against official databases — AI-hallucinated citations give you independent grounds to challenge the entire assessment, regardless of the underlying merits.
• If appearing on behalf of the Revenue, review AI-assisted legal research thoroughly before incorporating any citation into an official order or submission. A quashed order does not just harm the taxpayer — it exposes the department to costs and damages departmental credibility.
• For tax practitioners: document your citation verification process. In the event an AI-assisted submission is later challenged, a verified record of your research process is your best defence.

— — —

---

DEEP DIVE: How an AI Hallucination Created — and Destroyed — a Rs. 27.91 Crore Tax Demand

The Assessment and the Fatal Citations

KMG Wires Private Limited faced what appeared to be a substantial income tax addition: Rs. 27.91 Crores. For a manufacturing company, this is not an abstract sum — it can determine whether a business survives the financial year, whether credit facilities remain intact, and whether directors face personal liability.

The Assessing Officer’s order contained what appeared, at first glance, to be a thoroughly reasoned legal basis: multiple judicial precedents supporting the position taken. The citations were formatted correctly, with case names, court names, and year references that looked entirely plausible. A casual reader — or a busy tax practitioner under time pressure — might have assumed they were genuine.

They were not. When KMG Wires’ counsel checked each citation against Manupatra, SCC Online, and IndianKanoon, none could be found. Cross-referencing with official court databases confirmed the same result: the judgments did not exist anywhere in the Indian judicial record.

The Writ Petition and the Bombay HC’s Response

KMG Wires filed a writ petition under Article 226 before the Bombay High Court. The challenge was not initially framed as an ‘AI case’ — it was framed as a challenge to an assessment order that cited non-existent legal authority. Justices Colabawalla and Khata, hearing the matter, quickly appreciated the gravity of what was before them.

The court’s reasoning was clear and direct. A quasi-judicial authority — the Assessing Officer — exercises statutory power that directly affects a citizen’s fundamental rights, including the right to property and the right to carry on business. When that authority cites fabricated legal precedents as the basis for its decision, it acts without legal authority. There is no ‘honest mistake’ exception. There is no ‘good faith AI use’ defence. The absence of real legal authority renders the order a nullity.

The court quashed the assessment in its entirety and remitted the matter for fresh adjudication — instructing the AO to comply with the principles of natural justice and to base any fresh assessment on actual, verifiable legal authority. The subtext was unmistakeable: use AI at your peril; verify everything.

The Broader Income Tax System Risk

India’s income tax system processes tens of millions of assessments annually. The Central Board of Direct Taxes (CBDT) has been actively encouraging the use of technology by AOs. If AOs across the country are using AI tools to assist in drafting assessment orders — and incorporating AI-generated citations without verification — the scale of potential illegality is staggering.

Every assessment order that cites a non-existent judgment is potentially void. Every taxpayer who has paid tax pursuant to such an order may have a claim for refund. And every AO who has signed such an order may have created personal exposure.

The KMG Wires judgment is therefore not just significant for the company — it is a warning shot across the entire income tax administration. The CBDT would be well-advised to issue an immediate circular prohibiting the use of AI-generated citations in assessment orders unless independently verified, and mandating that AOs document their citation verification process.

Comparison with Global AI Hallucination Cases in Tax

While AI hallucination cases in legal proceedings have been documented globally, cases specifically in tax administration are rarer. The US Internal Revenue Service (IRS) has been cautious about AI adoption in formal proceedings. The UK’s HMRC has piloted AI for compliance but has not reported documented hallucination incidents in formal assessments.

India’s case is notable precisely because it has produced a clear, published judicial ruling: that AI-hallucinated citations in a tax assessment void the assessment. This gives Indian taxpayers a powerful tool to challenge assessments where AI use is suspected — and gives Indian courts a clear precedent to follow in future cases.

What to Watch For

The Income Tax Department’s response to this judgment will be telling. If the department issues guidance requiring AOs to verify citations before incorporation into orders, it would signal institutional acknowledgment of the AI risk. If no guidance follows, advocates and taxpayers should treat every assessment order as a potential candidate for the ‘citation check’ — a quick verification that takes minutes but can save a multi-crore dispute.

The ITAT is also likely to be confronted with this issue — indeed, the Buckeye Trust ITAT judgment (30 December 2024) appears to have involved the Tribunal itself using AI-generated content, with the order subsequently being withdrawn. The institutional consequences of AI misuse are spreading upward through the appellate hierarchy.