Guidelines for Examination of Computer Related Inventions (CRIs), 2025

Section 3(k) of the Patents Act, 1970 excludes from patentability "a mathematical or business method or a computer programme per se or algorithms". The present wording came with the Patents (Amendment) Act, 2002, which took effect on 20 May 2003 and added the words "per se". The Guidelines record that a Joint Parliamentary Committee explained the addition: computer programmes may include other things ancillary to them, and the intention was not to reject them for grant if they are inventions, but only to deny patents to computer programmes "as such". They identify four limbs of the exclusion: mathematical method, business method, algorithm and computer programme per se.¹

The Indian Patent Office has issued guidance on computer related inventions (CRIs) several times. Its guidelines page lists a draft in 2013, guidelines in 2016 and revised guidelines in 2017, all of which remain on the website beside the 2025 text.² SpicyIP's commentary describes the 2025 version as the first update of the CRI guidelines after about eight years and the fourth such iteration.⁴⁶

The Guidelines rely on a line of High Court decisions, in particular Ferid Allani v. Union of India (Delhi High Court, 12 December 2019), which said that patent offices worldwide test computer-related applications on the "technical effect" and "technical contribution" and that an invention showing either is patentable even if based on a computer program, as well as Microsoft Technology Licensing LLC v. Assistant Controller of Patents and Designs (Madras High Court, 3 July 2024) on the meaning of "algorithm" and "per se", and decisions such as OpenTV, Raytheon and BlackBerry.¹⁴

The Patent Office released the Draft Guidelines for Examination of CRIs, 2025 for public consultation on 25 March 2025, initially seeking comments by 15 April 2025. The draft added a discussion of recent case law and illustrative examples of patentable and non-patentable claims.³ SpicyIP's commentary said the consultation window was too short, and recorded that the Patent Office later extended the deadline to 30 April 2025.⁴ The Patent Office's website lists the comments and feedback received on "Draft Versions 1.0 and 2.0" and details of an in-person consultation, so the draft went through at least two versions before the final text.²

The final Guidelines are published on the IP Office website with two annexures: Annexure I of additional illustrative examples and Annexure II, a non-exhaustive list of case laws.² Commentary refers to them as the July 2025 CRI Guidelines.⁶ The text states that they apply with immediate effect and that the provisions on Section 3(k) in Chapter 09.03.05.10 of the Manual of Patent Office Practice and Procedure stand deleted.¹

The Guidelines have seven parts: an introduction; terms and definitions; legal provisions and recent jurisprudence; the examination procedure (novelty, inventive step, industrial applicability, sufficiency of disclosure and the determination of excluded subject matter); a chapter on AI, ML and DL, blockchain and quantum computing; a saving clause; and applicability.¹

Their stated aim is to foster consistency in examination and to bring clarity on the exclusions in Section 3(k) so that eligible CRI applications can be examined efficiently. They stress that the examiner should judge the substance of the claims taken as a whole, whatever their form (method, system or computer programme product), and that a claim is excluded only if in substance it falls within the excluded categories.¹

For each limb of Section 3(k) the Guidelines prescribe steps. For algorithms, the examiner identifies the series of steps in the claim and then asks whether they are abstract, devoid of the technical specifics needed to implement them, or are enabled with the technical components that produce a technical solution to a real-world problem; the first is excluded and the second is not. For computer programmes per se, the examiner construes the substance of the claim, identifies the core problem and solution and the essential technical features, and asks whether the technicality produces a technical effect "beyond a mere incidental effect"; if so, the claim is not excluded.¹

SpicyIP commentators criticised the draft for introducing overlapping terms (technical effect, technical contribution, technical advance and inventive contribution), for importing a "technical advance" step into inventive-step analysis, and for treating the different limbs of Section 3(k) in a way that left open whether computer programmes are tested more strictly than mathematical or business methods.⁴

Chapter 5 is the Patent Office's first systematic treatment of AI inventions. It begins by saying that AI, ML, deep learning, blockchain and quantum computing "harness computational architectures and algorithmic techniques" but are built on mathematical models and abstract principles, so examination raises particular challenges for sufficiency of disclosure and for exclusion under Section 3(k). The core question is whether the claimed invention uses theoretical constructs to provide an implementable technical solution or stays in the realm of abstract theory.¹

The Guidelines list the AI/ML constructs covered (neural networks including CNNs, RNNs, spiking networks, GANs and transformers, and classical methods such as random forests, decision trees and Bayesian networks, with mechanisms such as backpropagation and attention) and state that these are inherently abstract mathematical entities. "An abstract idea", such as a mathematical formula or a theoretical construct of AI, ML or DL, is not patentable, but an AI innovation that transforms abstract principles into a real-world tangible application may become patentable subject matter.¹

Inventorship (5.2.1). The Guidelines divide AI-related inventions into two groups. AI-generated inventions, created by AI systems autonomously or with very limited human intervention, are "not patentable as AI cannot be termed as a 'Person'" claiming to be the true and first inventor under Section 6 of the Patents Act. AI-assisted inventions, made with AI as a tool in the inventive process, are not categorically non-patentable under Section 3(k) if they meet the patentability criteria and demonstrate technical effect through tangible inventive applications.¹ This is the Patent Office's position on the question raised by the DABUS applications worldwide, and it parallels the Copyright Office's reported refusal in August 2026 to recognise an AI system as an author (see the entry on the Copyright Act, 1957 and AI).

Sufficiency of disclosure (5.2.2). The Guidelines warn that the theoretical, mathematical and algorithmic nature of AI inventions can allow vision statements and extrapolated use cases to be "camouflaged" as solutions without specific detail. They quote the Madras High Court in Caleb Suresh Motupalli v. Controller of Patents (C.M.A. (PT) No. 2 of 2024, 29 January 2025), which held that a claimed AI-related persona-extension invention failed the enablement requirement under Section 10(4)(a) and the best-mode requirement under Section 10(4)(b), and which also discussed the clarity and fair-basing rule in Section 10(5). The Guidelines conclude that disclosure must enable a person skilled in the art to reproduce the invention without undue experimentation.¹

Illustrative scenarios (5.2.3). Seven scenarios and Examples 9 to 16 show the disclosure expected. They include: where the claim covers an input-to-output transformation, the specification should explain the transformation and give test evidence of accuracy and consistency; where it relies on a trained model, the training data, architecture and training methodology and validation must be described; where pre-processing is key, every pre-processing step and its link to the model must be disclosed; for reinforcement learning, the agent, states, actions and rewards; where the claim improves the computer's internal operation, how the algorithm interacts with the hardware; where the novelty lies in dataset traits, those traits; and, depending on whether the invention lies in a new learning algorithm or in data engineering, the full architecture and training protocol or the data pipeline and its rationale. The hypothetical examples include digitising handwritten prescriptions with a CNN and transformer, failure prediction with LSTMs, crop classification from satellite images, a reinforcement-learning traffic signal system, a neural-network task scheduler for an operating system kernel and facial recognition tuned for elderly subjects.¹

SpicyIP's analysis of the draft concluded that it added case law and examples but did not resolve how Section 3(k) interacts with inventive step, and that it "raises a lot of questions". It also noted that the draft cited OpenTV as treating the bar on business methods as absolute but omitted decisions taking an implementation-focused approach.⁴

The final Guidelines were considered in a Calcutta High Court decision, Google LLC v. The Controller of Patents, of 6 August 2025. SpicyIP reported that the Controller had relied on the 2017 test and that the court upheld the rejection of an invention running on standard software with no technical contribution, because the appeal arose from a 2020 order, leaving it unclear how far the court relied on the 2025 text.⁵

In May 2026 a SpicyIP commentator argued that despite the July 2025 Guidelines, Section 3(k) "remains unclear", calling it a "hot mess" in which the Delhi High Court has applied six tests built on the word "technical" and Delhi and Madras decisions are inconsistent on similar facts. The post discussed a Delhi High Court decision in BlackBerry v. Controller in early May 2026 that rejected a method of colour-coding message recipients as unpatentable under Section 3(k), the court holding that an error in sending a message is not a technical problem.⁶ The commentary reviewed does not discuss the AI chapter specifically.

As of 15 October 2026 the 2025 CRI Guidelines are listed on the Patent Office's guidelines page beside the 2017 revised guidelines and earlier versions, with Annexures I and II. The page does not state that the 2017 text is formally superseded, but the 2025 Guidelines delete the Manual provisions on Section 3(k) and apply with immediate effect.¹²

On 7 August 2026 the Patent Office also published its Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures, which are about the Office's own use of AI tools and are not part of the CRI Guidelines. In October 2026 the Office also released draft examination guidelines for pharmaceutical and biotechnology applications.²

The Guidelines remain examiner guidance. They do not bind courts, which continue to develop the Section 3(k) jurisprudence, and the sources reviewed identified no Indian court decision recognising an AI system as an inventor.