India's Trade Marks Registry Drafts AI Rules for Its Filings -- an AI-Use Declaration With Submissions, No AI in Affidavits or Evidence, and Native Files on Demand

On October 16, 2026, the Office of the Controller General of Patents, Designs and Trade Marks (CGPDTM) issued a public notice inviting "suggestions/objections" on comprehensive guidelines on "the use/misuse of Artificial Intelligence (AI) and the menace of squatting", in compliance with a Delhi High Court order of October 13 in Stryten Energy LLC v. Nidhi Agarwal & Ors. Comments go to llc-ipo@gov.in "within a period of three (3) weeks" of publication, and the notice says the guidelines "shall be implemented for all pending applications and the new applications received by the office." MediaNama, reporting on October 19, also published a separate 23-page draft, "Guidelines on Use of Artificial Intelligence in Proceedings before the Trade Marks Registry", prepared under an earlier order in the same case dated April 16. The draft addresses AI only, not squatting, and is not yet in force.

What the draft requires

• Disclosure. Clause 5.2: "The obligation of disclosure is absolute. Failure to disclose AI use when it has occurred shall be treated as a misrepresentation to the Registry". AI is defined broadly to include large language models, generative AI systems and "Natural Language Processing tools", borrowing the Gujarat High Court's 2026 court-AI policy definition.
• Verification. Every AI-contributed citation must be checked against a non-AI legal source, and a non-existent or misleading citation is a ground for action "regardless of whether such incorrect citation originated from an AI".
• Sworn statements. Clause 14.1 bars using AI to "generate, author, embellish, strengthen, dilute, fabricate, or otherwise alter the substantive content" of affidavits and declarations.
• Evidence. Clause 14.2 bars AI that would "generate, fabricate, materially manipulate, or digitally alter" labels, invoices, sales data, market surveys, website extracts, advertisements, translations and proof of use. The Registrar may demand "the original or native file" plus metadata, audit trails and server logs.
• Penalties. After a hearing the Registrar may strike off a citation or the whole submission, and refer trade-mark agents to a disciplinary committee or advocates to their Bar Council. Officers must keep AI "strictly ancillary" to statutory decisions.

What remains uncertain. The draft contradicts itself on scope: clause 8.1 requires a declaration where AI was used, while clause 8.3 requires one with every submission "stating whether or not the submission has been prepared with the assistance of an AI". MediaNama notes the Annexure I form has no field for that answer, and has asked whether grammar checkers or machine translation trigger disclosure, and whether a real label designed with an AI tool can be filed as evidence of use. The draft also says it will be read with "Supreme Court Artificial Intelligence Regulations, 2026" that MediaNama could not confirm have been notified. Penalties fall on parties and representatives, not on officers who rely on unverified AI output.

Analysis: This is one of the first Indian administrative bodies to propose AI rules at the level of individual filings, and the design choices will travel: a per-submission declaration, personal liability for citations, and a power to demand native files that only works if firms keep them. The broad definition is the weak point. If spell-check and translation software count as AI, nearly every filing must declare, and the declaration loses meaning. For law firms and brand owners the practical step is the same either way -- keep originals and creation metadata for every label, advertisement and invoice filed as evidence, and log which tools touched each draft -- before the three-week comment window closes in early November.