
Introduction
With the use of artificial intelligence (AI) permeating into legal practice, and particularly, the management of intellectual property (IP) portfolios, a host of risks relating to AI hallucinations have followed, such as systems inventing non-existent case laws, misstating operative parts of decisions, or even filing incorrect facts.
After directions from the Delhi High Court in 2026 [1], the Office of the Controller General of Patents, Designs and Trade Marks (CGPDTM), has issued draft Guidelines on Use of Artificial Intelligence in Proceedings before the Trade Marks Registry, 2026 (“Draft Guidelines”), seeking to address some emerging issues. These were open for comments until recently.
Summary of the Draft Guidelines
The primary objectives of the Draft Guidelines are to:
- Prevent the use, misuse, and abuse of AI-generated content, including fabricated, false, or non-existent case law citations, in proceedings before the Trade Marks Registry (TMR);
- Ensure transparency, disclosure, and human accountability for all AI-assisted submissions placed before the Registrar;
- Establish a framework for verification, and consequences of non-compliance consistent with the TM Act, 1999, and TM Rules, 2017;
- Protect the integrity of quasi-judicial proceedings before the Registry and ensure that decisions of the Registrar are based on verified, accurate, and authentic legal and factual material; and
- Provide a deterrent against perjury and obstruction of justice arising from unverified AI-generated content.
The Draft Guidelines offer key definitions for terms such as AI, AI-generated content, generative AI, hallucination, and human-in-the-loop. These appear to have been borrowed mostly from the Policy on the Use of Artificial Intelligence in Judicial and Court Administration, 2026, issued by the High Court of Gujarat. The Draft Guidelines are expected to operate in addition to, and not in derogation of, legislation already in place (with the laws taking precedence over the draft guidelines), specifically, the TM Act, the TM Rules, the Bharatiya Sakshya Adhiniyam, 2023 (the evidence code), the Information Technology Act, 2000; the Digital Personal Data Protection Act, 2023; and any other law in force on the admissibility and appreciation of evidence or the regulation of digital content and personal data.
The Draft Guidelines focus on a “Human-in-the-Loop”, requiring that every submission filed before the TMR must be accompanied by a standardized declaration as to whether AI was used or not. If this is in the affirmative, the declaration must confirm that all cited precedents, provisions, and facts were independently cross-checked using “Authorised and Recognised Non-AI Legal Sources” (e.g., official court / TMR websites, and Indian law reports and databases).
The Draft Guidelines do not provide any opportunity for disclaiming responsibility. In other words, a party or representative cannot forego its submission responsibilities by blaming a false or fabricated submission on an AI tool.
Any party, representative, or officer who detects a fabricated citation or misrepresentation can report the same to the Registrar. While good-faith whistleblowers are protected from adverse consequences, filing frivolous or vexatious reports can attract penalties.
If the Registrar suspects that a document lacks authenticity due to AI manipulation, the Draft Guidelines empower them to invoke Section 127 of the TM Act, and initiate proceedings, where, after a show-cause notice and hearing, the false citation may be struck off, or the entire submission or document dismissed. For professional misconduct or perjury, the Registrar can refer trade mark agents to the competent disciplinary committee (currently, the Ad hoc Disciplinary Committee for Trade Mark Agents) and advocates to the respective State Bar Councils, for appropriate disciplinary proceedings.
Some concerns
The Draft Guidelines are arguably well-intentioned, but present many concerns.
The highest order question perhaps requires asking whether it is the role of regulators and quasi-judicial authorities to separately issue such advisories and guidelines at all. If each such entity issues separate guidelines, it opens up issues of parity of treatment across different regulatory frameworks. For example, within the intellectual property arena alone, these Draft Guidelines pertain only to trademarks, with no indication as to whether they will apply also to submissions made to other offices or registries, such as the Patent Office or the Design Registry.
The second order question relates to the extent of authorisation delegated to the CGPDTM in the first place. While it is true that the Delhi High Court directed the CGPDTM to draft such guidelines, the Draft Guidelines may not be the appropriate vehicle for imposing definitional restrictions on key terms such as “Hallucinatory AI”, or for directing penal consequences.
The third order question relates to the distinction in different types of AI use. It would be safe to say that some form of administrative automation in the filing of applications is here to stay, particularly for routine work, such as proofreading, formatting, translation, or transcription. To treat these activities the same as using AI that creates hallucinatory content is concerning, especially with declaratory requirements being so onerous, and corresponding penalties being reasonably harsh (striking off the mark, disciplinary action, etc.). Some form of accountability is undoubtedly needed, but perhaps a more nuanced treatment of AI use would be better suited for such a framework.
Relatedly, minor implementation niggles need to be ironed out, such as through providing guidance and illustrative examples to clarify the intended scope of the requirements. A reasonable transition period may also be considered to enable practitioners and other stakeholders to adapt their filing processes, if such requirements are to be finally implemented.
Separately, a question arises as to the accountability around AI use by the TM Registry itself. Here, it is useful to consider that the Draft Guidelines are intended as an outward-facing document, directed at external actors, such as parties, TM agents, and advocates appearing before the TMR. Contrastingly, earlier in 2026, the Patent Office had issued draft Patent AI Guidelines, which was meant as an internal policy for the Patent Office itself, governing patent screening, prior-art search, and examination (seeking mainly to improve examination quality while protecting patent applications from public AI models).) This invites a scenario of asymmetric accountability. There is no indication as to whether officials of the TMR will be similarly held accountable to standards of the quality of examination and adjudicatory decisions.
A consideration of all of these issues will surely help clarify the intention behind the Draft Guidelines better, and improve the quality of regulation overall. Ideally, a future version of this document (or a document like this) should look towards standardizing protocols across registries within the CGPDTM, aligning internal expectations with external compliance, and ensuring uniform digital evidence checks.
References:
[1] Stryten Energy LLC v. Nidhi Agarwal & Ors. (Order of the Delhi High Court dated 16 April 2026)













