The draft Guidelines on the Use of Artificial Intelligence in Proceedings before the Trade Marks Registry seek to address the growing concerns around AI-generated fabricated citations and misleading evidence. However, as explained by Kaustub Boidi, the proposed framework leaves important questions unanswered about the Registry’s own verification responsibilities, the treatment of legitimate AI-assisted translations, and whether the prescribed consequences are proportionate to different forms of non-compliance. Kaustub is a final-year law student at SVKM’s NMIMS, Hyderabad. His interests lie in intellectual property law, particularly trademarks, and the evolving intersection of artificial intelligence and law.
A Fake Judgment, an AI-Designed Mark and a Missing Declaration: What Do the TMR’s Draft AI Guidelines Actually Regulate?
By Kaustub Boidi
An AI-generated fake judgment cited in a trademark opposition can mislead the Registrar about the law. In contrast, the use of AI to locate potentially relevant authorities that are subsequently checked against the judgments or the translation of a genuine invoice without altering the content is legitimate assistance. The recently published draft Guidelines on the Use of Artificial Intelligence in Proceedings before the Trade Marks Registry give rise to the question whether the rules are sufficient to distinguish such uses from AI that fabricates or materially misrepresents legal or factual material.
Following the Delhi High Court’s intervention in Stryten Energy LLC v. Nidhi Agarwal & Ors. (pdf), the CGPDTM proposed these draft Guidelines to address the use and misuse of AI in submissions and representations before the Trade Marks Registry, particularly to address the issues of fabricated case-law citations and misleading factual material. The draft includes guidelines governing the disclosure of AI use and verification of legal and factual content, as well as limitations on AI-generated content in certain evidentiary documents. The Guidelines are not limited to pleadings but are also applicable to applications, affidavits, declarations, statements and exhibits, images, labels, translations, invoices, advertising, websites extracts, market surveys, sales data and other documents filed in the course of examination, opposition, rectification and other proceedings before the Registry. The questions this post will explore are: How will the Registry verify checks declared by filers, can the relevant guidelines easily distinguish between fabrication and AI-assisted translation of genuine material, and do the proposed consequences match the various types of non-compliance?
How is the Verification Verified?
The guidelines mandate that the use of AI in preparing submissions before the Registry must be disclosed. The disclosure requirements are given under clauses 5 (Transparency and disclosure) and 8 (Mandatory declaration of AI use), with clause 5 acting as the general principles warranting the disclosure of AI use and clause 8 specifying how the disclosure is supposed to be made. Under clause 8, all submissions must include a declaration on the use of AI (Clause 8.3), and the declarant must ensure that it has verified the accuracy of the contents and that it accepts full personal responsibility for the accuracy and authenticity of the contents (Clause 8.2). The prescribed declaration form is given in Annexure I.
Clauses 6.1–6.3 further specify how the above verification will take place. They impose a requirement to independently verify information contributed by AI against a recognised non-AI source, factual assertions against the underlying record, document or personal knowledge, and to certify those verifications.
What about the Verification at the Registry’s End?
What happens at the Registry’s end is the more important question. Clause 11.2 requires Registry officers to independently verify AI-generated legal propositions, factual statements and precedents before relying on them in their official functions. Clauses 12 and 13 also enable the Registrar to initiate proceedings or to consider reports if there is a suspicion of false citations or misleading facts or declarations. However, the draft does not provide details on how the Registry will verify whether a filer has actually performed the verification certified in Annexure I. Does the declaration undergo scrutiny at the time of filing or does further scrutiny normally only start if a problem is suspected and/or reported? This problem is there in the recently published Draft Trademark Manual, where the Manual only talks about consent of the natural person whose AI-generated representation has been submitted for trademark registration but does not talk about how these representations and submissions are to be verified by the Registry.
It is to be remembered that these guidelines are for assisting the Examiners as well and while the guidelines talk about the consequences and the procedure for initiating proceedings when AI misuse is flagged, there seems to be an important gap of how the concerned officer is supposed to find the misuse. The draft Guidelines need not require every filing to undergo a fresh check of every citation. It should, however, differentiate the filer’s certification from the Registry’s verification and make clear how the latter is supposed to function.
Does Clause 14.2 Address AI Translations?
Clause 14.2 focuses on the prohibition of AI use to “generate, fabricate, materially manipulate, or digitally alter” a broad spectrum of material, such as images, labels, translations, advertisements and proof of use filed in support of proceedings. An explicit reference to “translation” in Clause 14.2 raises a more textually-based question. On a literal interpretation, the provision would apply to the use of AI to translate a true document, even if the original document is not touched and the translated version is reviewed for accuracy. The Supreme Court’s draft Regulations for Use of AI in Courts, 2026 clearly provide for the use of AI tools to translate legal documents, with the requirement for human verification of accuracy and fidelity to the original. While the two drafts are set in different contexts, the comparison emphasizes the importance of differentiating between fabricating evidence and translating an authentic document. The Registry thus should clarify whether 14.2 prohibits all types of AI-assisted translations or only those that fabricate or materially distort/alter the source.
When Does an Error Become Misconduct?
There is also a need for a more explicit criterion to guide action when a citation is wrong. Clause 9.1 requires verification of a case’s existence, citation details, attributed holding, and relevance. The proviso prevents action on the basis of disputed relevance of a citation, but mentions deliberate or reckless misattribution for determining when action may be taken.
Clause 9.3, however, provides that citing a non-existent case or misstating the holding of a cited case is sufficient grounds to take action under Clause 12, whether the error is caused by an AI system or not. Similarly, Clause 12.1(a) allows proceedings when a false or fabricated citation has been used, but does not specifically state an intent requirement.
This lack of certainty is significant. A deliberately invented judgment, a reckless misstatement of a real judgment and an inadvertent citation error are not necessarily equivalent forms of misconduct. The proviso to Clause 9.1 does seem to appreciate that distinction, but the connection between the proviso and Clauses 9.3 and 12.1 is not entirely clear. The draft should clarify that some breaches of the regulations entail deliberate or reckless conduct, while others may warrant action even if not intended.
This is no excuse to overlook faulty citations. It is an argument that the standard for correcting an error must not be mixed up with the standard for imposing disciplinary action. Unless there is a clear distinction, representatives will not be able to easily determine when a mistake can become actionable misconduct.
The Sanctions Do Not Match the Offences
The enforcement provisions expose a more concrete problem. There are four grounds for proceeding under Clause 12.1: false or fabricated citations, failure to disclose the use of AI, false declarations, and misleading or fabricated factual material created using AI. But under Clause 12.3, there are only two possible remedies: disregard of the offending citation or striking off of the whole submission containing the citation.
Consider a filing in which AI was used without disclosure, but every citation is genuine and every factual assertion is accurate. While Clause 12.1(b) expressly allows proceedings for non-disclosure, there is no clear wording in Clause 12.3 to indicate what happens where there is no offending citation to disregard or a submission containing the offending citation to strike off. The same ambivalence applies if the charge is against the truthful information rather than an invented source of authority.
In Clause 12.4, disciplinary referrals are permitted, but are stated as supplemental to the action under Clause 12.3. The draft does not explicitly clarify that in cases of non-disclosure or incorrect factual information, the only consequence could be a referral. Nor does it define when a procedural failure or ignoring specific content would be sufficient, and when it would be appropriate to either strike off an entire submission or to send a representative for disciplinary action. These differences should be clearly communicated, not implied by provisions based on false citations.
There is an associated issue with the reporting mechanism. Clause 13.6 makes a person who files a frivolous or vexatious report liable to action under Clause 12, even though Clause 12.1 does not expressly identify making such a report as a ground for proceedings. There are also some incorrect cross-references on clauses 13.2 and 13.3 regarding reports to be submitted and the time for production of supporting material. These are not just cosmetic errors. They influence the investigation of allegations and the parties’ exposure to consequences.
The draft should outline what action may be taken for each infraction and when it would be appropriate to correct an error or to ignore specific content, and when it would be appropriate to remove the whole submission from the record or refer a representative for disciplinary action. Show cause notice and opportunity to be heard are important safeguards but do not resolve uncertainty of grounds for each consequence.
The draft is an opportunity to establish meaningful safeguards against the misuse of AI in trademark proceedings. But requiring filers to declare and verify their use of AI is only part of the task. The Registry should also clarify how it will determine if the checks have been performed, and the restrictions should clearly differentiate between fabrication and legitimate assistance, while each enforcement provision should link a violation to a clear consequence. If these distinctions were absent, the Guidelines would run the risk of placing specific obligations on filers without commensurate clarity regarding the assessment of those obligations and enforcement.
