Indian Patent Office Guidance for Examining AI inventions- CRI Guidelines 2025

In recent years, the landscape of innovation has been profoundly shaped by advancements in computer-related technologies. As these innovations become increasingly sophisticated, the demand for clear and reliable patent protection in this domain has grown exponentially.  

Developers, inventors, and other stakeholders engaged in Computer Related Inventions (hereinafter referred to as CRIs) continuously seek appropriate means of protecting their innovations and investments—often turning to intellectual property frameworks, such as patent law, as a possible route. However, the dynamic and rapidly evolving nature of software- and algorithm-based technologies presents distinct challenges for the India Patent Office. These challenges have become even more pronounced with the exponential progress in artificial intelligence in recent years, further complicating the distinction between abstract concepts and patentable technical solutions. 

India, recognizing the growing importance of this technological sector, has taken a significant step forward. The Office of the Controller General of Patents, Designs & Trade Marks (CGPDTM), has released the Revised Guidelines for Examination of Computer Related Inventions, 2025.  Moreover, the IPO has announced in a public notice that the comments received during the revision of version 2.0 and the feedback from the stakeholders meeting will be released shortly. 

Having said the above, these updated guidelines aim to provide clarity and consistency in the examination of CRI patent applications, especially with regard to the interpretation and application of Article 3(k) of the Indian Patents Act. The primary aim of the new guidelines is to enhance legal certainty and transparency for stakeholders in the CRI ecosystem.  

In this regard, the mentioned guidelines acknowledge the complexity of inventions involving Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL), which often rely on advanced architectures such as neural networks and traditional ML methods like decision trees or Bayesian networks. These systems, built on mathematical and algorithmic constructs, are inherently abstract and not patentable in themselves accordingly to the current regulation. 

However, the guidelines clarify that AI/ML/DL inventions may be patentable when they result in tangible, real-world applications that demonstrate a technical effect—moving beyond abstract ideas into practical implementation. 

Two categories of AI-related inventorship are also addressed: 

• AI-generated inventions (created autonomously by AI): Not patentable, as AI cannot be recognized as an inventor under Indian law.
• AI-assisted inventions (where AI is used as a tool): Potentially patentable if they meet existing patentability requirements and offer a clear technical contribution. 

The guidelines also emphasize the importance of sufficient disclosure. Given the often-theoretical nature of AI-based inventions, applicants must avoid vague, speculative claims and ensure that concrete, detailed implementations are described, rather than hypothetical or overly broad use cases. 

In practice, applicants, particularly those from small and medium-sized enterprises (SMEs), must adapt the drafting of their AI/ML/DL-related patent applications to suit the specific nature of their invention. The revised guidelines provide a series of illustrative, non-exhaustive scenarios that serve as a reference for understanding how patentability may be argued in various technical contexts. 

For instance, in cases where the invention primarily involves a complex transformation between known inputs and outputs—without fully disclosing the internal logic of the model—the specification should clearly explain how this transformation is achieved. This includes detailing known processes, parameters, and variables to ensure that a person skilled in the art can reproduce the outcome. Fulfilling the sufficiency of disclosure requirement in such scenarios is crucial to ensure that the invention is not dismissed as a mere abstract idea without a concrete technical application. 

When the claimed invention is based on an already trained AI model, the focus should shift to demonstrating a clear relationship between the training data and the technical problem being addressed. This means identifying the dataset used, explaining its relevance to the problem, and showing how it enables the model to produce effective and reliable results. A comprehensive description of the model architecture, training methodology, and parameters is also necessary, supported by validation metrics and real-world test outcomes. These elements collectively help establish that the invention offers a concrete technical solution and is not merely leveraging generic AI techniques. 

In situations where the invention claims to improve the internal functioning or structure of a computer system, the application should describe in detail how the underlying algorithm interacts with the hardware or system architecture to produce a measurable technical benefit. Likewise, if the inventive step lies in the unique characteristics of the training data—such as its origin, diversity, labelling methodology, or specific statistical traits—these must be disclosed, unless they are readily apparent to a person skilled in the art. In most cases, it is not necessary to include the entire dataset, especially if it is large or sensitive, but its defining features should be clearly described to illustrate why it is critical to the model’s performance and problem-solving capability. 

The guidelines also touch upon emerging fields such as quantum computing, which, while grounded in abstract physical principles like superposition and entanglement, may become patentable when those principles are applied in a concrete, technical manner. For example, inventions involving novel hardware configurations for qubit manipulation or practical algorithms for quantum-enabled optimization can qualify as patentable subject matter—provided they demonstrate a tangible technical effect. As with AI, the focus remains on moving from theoretical constructs to real-world applications that offer measurable improvements in a technological process. 

Beyond AI and quantum technologies, the revised guidelines further extend their scope to address other cutting-edge domains such as blockchain, reinforcing the need for clarity, sufficiency of disclosure, and a demonstrable technical contribution across a broad spectrum of digital innovation. 

We hope that the publication of this guide will provide greater legal certainty when applying for this type of invention, given the controversy surrounding case law on computer-implemented inventions in recent years.