India's Trade Marks Registry Proposes Mandatory AI Declarations and Penalties for Fake Citations

By Dafa Attas

Photo: ds_30 / Pixabay

India's trade mark office wants anyone appearing before it to say whether artificial intelligence helped prepare their filings, and to check every case they cite against an official source. India's IP office, the CGPDTM, which runs the patent, design and trade mark registries, has released a 2026 draft set of guidelines on AI use in Registry proceedings. A public notice dated 16 October 2026 gives stakeholders three weeks to send comments to llc-ipo@gov.in, which puts the deadline in the first week of November.

The draft did not come out of nowhere. It responds to directions of the Delhi High Court in a writ petition filed by Stryten Energy LLC (W.P.(C)-IPD 16/2026), in which the court was confronted with AI-generated pleadings that cited judgments which do not exist. The court asked the Registry to propose guidelines on AI misuse and on trade mark squatting. Commentary on SpicyIP on 30 October questioned how much of the AI problem the draft actually reaches.

What the draft would require

A declaration on every filing. Each submission would carry a single declaration in a set form (Annexure I) stating whether AI was used. If it was not, saying so is enough. If it was, the declarant must confirm that every citation, statutory reference, legal proposition and material fact produced or suggested by the tool has been independently checked, and must accept personal responsibility for the result. Where a representative files, both the party and the representative sign.

Verification against non-AI sources. Checking means going to a recognised source without relying on AI output: the websites of the Supreme Court and High Courts, the Official Gazette, certified copies, established Indian law reports and databases such as SCC Online and Manupatra, or the Registry's own records. A citation an AI tool supplies does not count as verified until it has been cross-checked in one of those places.

Duties for examiners too. Registry officers would have to keep AI strictly in a supporting role and could not rely on an AI-generated legal proposition or precedent without the same independent check.

No AI-written sworn evidence. The draft would bar using AI to write or embellish the substance of affidavits and declarations, and to fabricate or alter evidence such as invoices, labels, sales data, survey results or website extracts. Where a document looks AI-manipulated, the Registrar could demand the native file, metadata or audit trails, exclude the material and draw adverse inferences.

Consequences

If the Registrar suspects a fake citation, a missing declaration, a false declaration or fabricated facts, it may act on its own or at a party's request. After a show-cause notice and a hearing, the Registrar could strike out the offending citation alone or the whole submission. Trade mark agents could be referred for discipline, and advocates referred to their Bar Council. Any party or officer could report a suspected fabrication on a set form (Annexure II). Good-faith reports are protected, but frivolous ones can themselves lead to action.

Two details deserve attention. First, the draft says that citing a non-existent case is a ground for action whether or not AI produced it. In practice it creates a general duty of accuracy, not just an AI rule. Second, a dispute over a citation's relevance alone would not trigger penalties; action lies only where a case does not exist or is deliberately or recklessly misstated.

Open questions

The draft is ambitious, and some parts may need work before it becomes final. The declaration is required on every filing, including routine replies to examination reports, so the Registry will face a large volume of paperwork. The definition of AI is broad and borrows from a High Court of Gujarat policy. It could cover ordinary tools such as translation or grammar software, so practitioners will want clarity on where the line sits. There is also a mismatch in dates to be resolved: the draft refers to a court order of 16 April 2026, while the public notice cites an order of 13 October 2026. Writing on SpicyIP, Kaustub Boidi points to two further gaps. The draft does not explain how the Registry will test whether a filer really carried out the checks it certifies. It also does not clearly separate fabrication from legitimate help, such as using AI to translate a genuine invoice without changing its content. Finally, the notice says the final guidelines would apply to pending as well as new applications.

An Indonesian angle

We are not aware of a DGIP regulation or an Indonesian court decision that deals specifically with AI-drafted submissions in trade mark proceedings. Indonesian practitioners should still take note. Opposition and appeal filings before DGIP and the Trademark Appeal Commission can rely on case law, and the risk of a hallucinated citation is the same in Bahasa Indonesia as in English. For foreign clients with portfolios in both India and Indonesia, the safe course is to apply India's proposed standard everywhere: check every authority against an official source and keep a record of that check.

Sources

• CGPDTM: Draft Guidelines on Use of Artificial Intelligence in Proceedings before the Trade Marks Registry (PDF)
• CGPDTM: Public notice inviting suggestions on AI and squatting guidelines, 16 October 2026 (PDF)
• SpicyIP: What do the TMR's draft AI guidelines actually regulate? (30 Oct 2026)
• IP India: news and updates