Comments and Suggestions on Draft CRI Guidelines 2025 issued by Indian Patent Office

BananaIP Counsels has submitted its comments and recommendation with respect to the Draft Guidelines for Examination of Computer Related Inventions (CRIs) issued by the Office of the Controller General of Patents, Designs, and Trade Marks. Through this post, we are sharing the submissions made to the Patent Office.

A) Background

The Office of the Controller General of Patents, Designs, and Trade Marks (hereinafter referred to as “Patent Office”) issued a public notice inviting comments and suggestions on the draft guidelines for examination of Computer Related Inventions (CRIs) (hereinafter referred to as “Draft CRI Guidelines”) on 15 April 2025. In furtherance of the said notice, IP attorneys and Patent Agents at BananaIP Counsels (“BananaIP”) are hereby submitting their views and suggestions with respect to the Draft CRI Guidelines for the consideration of the Patent Office.

B)  Comments and Suggestions

Overall, the guidelines cover the jurisprudence relating to CRIs in the form of computer programs and related inventions adequately, and no particular comments and suggestions have been made with respect to the same. The comments and suggestions in this submission particularly focus on portions of the draft guidelines relating to Machine Learning (ML), Artificial Intelligence (AI), and Blockchain patent application. Specific portions of the draft CRI guidelines have been provided for ease of reference before the comments and recommendations with respect to the said portions below. It may be noted that the comments and suggestions submitted particularly relate to sufficiency of disclosure guidelines.

PARTICULAR EXTRACTS, COMMENTS, AND RECOMMENDATION:

1) EXTRACT:

“4.4 SUFFICIENCY OF DISCLOSURE:

(a) In AI systems, while the inputs and outputs are typically known, the logic that transforms input into output may be complex or abstract. Description should aim to clarify this transformation as much as possible by detailing any known processes and variables. If test results or other forms of evidence validate the accuracy of the model’s output, these should be included, especially when the AI is used for precise applications where reliability is essential.”

COMMENT/SUGGESTION: –

The requirement to explicitly describe the transformation from input to output in AI systems may not always be practical, particularly when the model behavior is based on complex neural networks. In many cases, the internal logic may not be readily explainable, due to the inherent nature of machine learning systems.

RECOMMENDATION:

Flexibility may be allowed by permitting Applicants to show input-output correlation through high-level logic, representative examples, or statistical validation. Instead of insisting on detailed test results in every case, accepting performance metrics or model benchmarks, especially for commonly used models, would be a more practical approach.

2: EXTRACT

“(b) For a trained AI model, clearly defining the correlation between input and output data is critical. This correlation is considered fully described when:
i. The training data used for the model is explicitly identified,
ii. A link between the training data’s characteristics and the technical problem the invention addresses is made,
iii. The specific learning model and training methodology are comprehensively described, and
iv. The model, when trained, is shown to effectively address the technical problem with predictable results.”

COMMENT/SUGGESTION: –

Requiring full data disclosure may not be suitable in this scenario, particularly when dealing with large or proprietary datasets. It may also be noted that the training dataset, if disclosed in the patent application, does not have any other form of protection under the Indian law, and will form part of public domain, once the patent application is published. This may prove to be detrimental to the interests of AI innovation and patent applicants and may result in less patent filings because of high disclosure standards.

RECOMMENDATION: –

Alignment with the approaches of the EPO and USPTO may be considered, and the description of dataset characteristics, such as type, distribution, and relevance, may be permitted instead of requiring full datasets. For instance, in the Guidelines for Examination in the EPO, Part G, Chapter IV, it is stated that, “If the technical effect depends on particular characteristics of the training dataset used, the characteristics required to reproduce the technical effect must be disclosed unless the skilled person can determine them without undue burden using common general knowledge. However, in general, there is no need to disclose the specific training dataset itself.”

3) EXTRACT:

“(c) If data pre-processing plays a key role in the invention, all steps and functions of pre-processing should be disclosed, along with how they correlate to the end model. If this correlation isn’t clear or if a person skilled in the art might struggle to understand the link between raw data and processed learning data, the application risks failing to meet the enablement requirement.”

COMMENTS/SUGGESTION: –

Pre-processing often uses standard techniques, which may not require exhaustive disclosure unless they are innovative.

RECOMMENDATION: –

It is recommended that detailed disclosure be considered by the Patent Office only where pre-processing steps contribute to the inventive concept or technical effect; otherwise, referencing known methods may be deemed sufficient.

4) EXTRACT

“(d) For AI applications utilizing reinforcement learning, the application must specify how the system interacts with its environment, including agent interactions, states, actions, and rewards. Omitting these details, or failing to describe them in a way that a person skilled in the art can deduce, could result in a nonenabling disclosure.”

COMMENTS/SUGGESTION: –

This requirement may be reasonable for inventions claiming Reinforcement Learning (RL) behavior, but not all RL-related inventions might definitely have agents, states, actions, and rewards.

RECOMMENDATION: –

The disclosure of agents, states, actions, and rewards with respect to reinforcement learning (RL) may be required only when those particular agents, states, actions, and rewards are essential to the technical effect or contribution. It may be noted that, in decision T 1952/21 of the EPO Boards of Appeal, the Board opined that the mere mention of reinforcement learning does not automatically confer technical character to an invention. The Board stated that “the concept of reinforcement learning in general does not imply a technical context” and clarified that a system using RL must contribute to the solution of a technical problem to be considered patentable. Therefore, when RL interactions, such as agent-environment setup, actions, and rewards, are essential to achieving the technical effect, they must be disclosed sufficiently.

5) EXTRACT:

“(f) When the invention’s technical effect depends on specific traits of the training dataset, these traits must be disclosed unless a person skilled in the art could identify them without undue experimentation. In most cases, it’s sufficient to describe the data’s defining characteristics rather than the specific dataset itself.”

COMMENTS/SUGGESTION: –

The Patent Office could clarify what is meant by “specific traits” of a training dataset, as the term may be interpreted differently depending on the context. Including examples such as data distribution, class balance, or feature relevance would enhance clarity.

6) EXTRACT:

“(g) Blockchain patent applications are required to include comprehensive descriptions of the cryptographic techniques used, the specific data structures involved, the consensus mechanisms employed, and any interactions with hardware or network systems. These detailed disclosures enable others to fully understand, replicate, and assess the functionality and innovation of the blockchain technology described. Blockchain patent applications must clearly define elements like distributed ledgers, consensus mechanisms, cryptographic processes, and network configurations. Clear descriptions of consensus mechanisms and data layouts (e.g., block structures, linkages) are crucial for enablement.”

COMMENTS/SUGGESTION: –

The requirement for detailed disclosure of cryptographic methods, data structures, and consensus mechanisms is reasonable and supports a clear understanding of blockchain-related inventions. However, some of these components are standard and well-documented in the public domain. Requiring disclosure of cryptographic methods, data structures, and consensus mechanisms may not be reasonable and may result in unjustified refusal of patent applications.

RECOMMENDATION: –

The Patent Office may allow Applicants to refer to established cryptographic protocols and consensus algorithms by reference, without reproducing their full technical details, where such references might be sufficient for an enabling disclosure. Also, even if such references are not provided, the Applicants may be permitted to rely on common general knowledge with respect of well-established and standard techniques, protocols, etc.

7) EXTRACT:

“(h) If the invention employs a novel machine learning technique, a comprehensive description is mandatory. This should cover essential aspects, such as the structure of neural networks, activation functions, network topology, convergence criteria, metadata and the learning mechanisms used. Each component of the algorithm should be disclosed to the extent it is necessary to achieve the invention’s claimed technical effects, ensuring that a person skilled in the art can replicate the process accurately.”

COMMENTS/SUGGESTION: –

The requirement of algorithmic details may result in undue burden on patent applicants.

RECOMMENDATION: –

The Patent Office may require structural or algorithmic disclosures to support functional claims only in specific instances where the invention lies in the configuration or training methodology. Otherwise, a general architecture-level disclosure may be considered to be sufficient to meet the disclosure requirements. This is consistent with the EPO’s position under Guidelines for Examination G-II, 3.3.1, which states that, “If the technical effect depends on particular characteristics of the training dataset used, the characteristics required to reproduce the technical effect must be disclosed unless the skilled person can determine them without undue burden using common general knowledge.”

C) Disclaimer

The comments, suggestions, and opinions provided in this document are based on authors of these comments at BananaIP Counsels. They may not be considered as generalization of any particular aspect or matter addressed in this document. It is understood that attorneys and experts within and outside BananaIP may have differing opinions, and that the suggestions provided are not the only ways of resolving issues expounded in the document.

The views expressed in this document do not reflect the views of BananaIP’s clients.

These comments, suggestions, and opinions with respect to the Draft CRI Guidelines have been submitted with the bona fide and honest intent of aiding the Patent Office.

Reviewed by: Dr. Kalyan Kankanala
Accessibility review by: Mr. Gaurav Mishra

You Might Also Want To Read

Author: Sowmya S Murthy

Sowmya S Murthy is an Indian Patent agent and a Managing Associate with BananaIP Counsels, a reputed IP firm and specializes in patent prosecution and drafting for electronics and software-related inventions. She also regularly contributes to blog posts on case laws and other topics related to patent practice. The views expressed in his articles and posts on Intellepedia are personal and do not represent those of BananaIP Counsels or its members. View all posts by Sowmya S Murthy