CRI explained: this guide covers what it means, who it applies to, the step-by-step process, documents required, fees, due dates and penalties in India — so you can stay compliant with confidence and avoid costly mistakes.
Paragraph 5 of the CRI Guidelines 2025 turns to artificial intelligence, machine learning and deep learning, with blockchain and quantum computing following. Paragraphs 5.1 to 5.2.2 cover the starting position for AI inventions, the question of who can be the inventor, and the disclosure the specification must make.
An AI, ML or DL model as an abstract idea or mathematical construct is not patentable; when it is turned into a real-world tangible application it may be (para 5.2). An invention created autonomously by AI is not patentable, because AI is not a "person" under section 6; an AI-assisted invention is not categorically excluded and needs a technical effect (para 5.2.1). The specification must disclose the invention fully and particularly, so that it can be reproduced without undue experimentation (para 5.2.2). The Guidelines are the Patent Office's guidance and do not have the force of law; the Patents Act, 1970 and the Patents Rules, 2003 as now in force prevail.
The Office revises its guidelines, so check the current version on ipindia.gov.in. This article is about patenting AI inventions. The separate guidelines on how the Office itself may use AI tools in examination are covered in our article on those guidelines. For help preparing an AI specification, see patent drafting and filing.
Why AI inventions get special treatment (paragraph 5.1)
The Guidelines observe that AI, ML and DL, like blockchain and quantum computing, rest on mathematical models, learning algorithms and abstract principles, and that their futuristic possibilities, realistic and unrealistic, appear in academic writing as possible use cases. They conclude that examination faces two unusual challenges: sufficiency of disclosure, and the section 3(k) exclusions.
The core question is stated this way: does the claimed invention use the theoretical constructs to give an implementable technical solution, or does it stay in abstract theory? Examination is therefore directed at separating a theoretical construct, resting on possibilities set out in the specification, from a practical application that solves a problem through technical means with enough detail for a skilled person to reproduce it without undue experimentation.
What the Office counts as AI, ML and DL subject matter (paragraph 5.2)
Paragraph 5.2 lists the families it has in mind: deep learning constructs (artificial, convolutional, recurrent and spiking neural networks, generative adversarial networks, transformers), traditional machine learning methods (random forests, decision trees, Bayesian networks, clustering and dimensionality reduction methods) and the mechanisms such as backpropagation, gradient descent and attention. Supervised, unsupervised and reinforcement approaches are included.
The Guidelines call these computational systems "inherently abstract mathematical entities" whose theoretical basis is independent of training. An abstract idea, such as a formula or a theoretical construct of AI, ML or DL, is not patentable because it lacks practical application. But when the innovation turns the abstract principles into a tangible real-world application, it may become patentable subject matter. How the exclusions are then applied is covered in the article on the four limbs and, for scenarios, the next article on paragraph 5.2.3.
Inventorship (paragraph 5.2.1)
The Guidelines divide AI-related inventions into two groups.
| Category | Meaning in the Guidelines | Position |
|---|---|---|
| AI-generated | Created by AI systems autonomously or with very limited human intervention | Not patentable, because AI cannot be a "person" claiming to be the true and first inventor under section 6 |
| AI-assisted | Made using AI as a tool in the inventive process | Not categorically non-patentable under section 3(k) if the patentability criteria are met and technical effect is shown through tangible inventive applications |
In practice, an applicant must name human inventors and be able to say what each human contributed. The AI system is a tool. For the statutory rule on who may apply, see our post on section 6.
Sufficiency of disclosure (paragraph 5.2.2)
The Guidelines warn that published future vision statements, the unrealised potential of the field and its mathematical nature can lead to prospective use cases being presented as solutions without specific details. Since patents are an exchange for disclosure, the claimed aspect must be disclosed fully and particularly.
The paragraph relies on a single ruling, Caleb Suresh Motupalli v Controller of Patents (Madras High Court, 29 January 2025), as cited in the Guidelines. In brief, the Guidelines draw four points from the excerpt.
- The court assumed the position of a notional person skilled in the art, who may be one expert or a team of experts, depending on the invention. For an AI invention that is a software engineer with AI expertise, or a team.
- The disclosure did not enable the claims: the specification described a concept and a set of metaphors, but not how to adopt the underlying techniques to achieve the result.
- A specification fails enablement where undue experimentation, needing inventive faculty, is required to work the invention.
- The same gaps meant the claims were not fairly based on the specification, and the court found the disclosure lacking under section 10(4)(a), (b) and (5).
The Guidelines conclude that the disclosure must enable reproducibility without undue experimentation by a person skilled in the art.
What the examiner checks in an AI specification
| Question | What the applicant should show |
|---|---|
| Is the model more than an abstract construct? | The application: what the model is applied to, what data it takes, what it outputs and how the output is used |
| Is the architecture described? | Layers, inputs, outputs, loss or objective, and training procedure in enough detail to reproduce |
| Is the data described? | Source, form and preparation of training and test data, as relevant |
| Is there evidence it works? | Results or working examples that show the claimed effect |
| Who is the inventor? | Named human inventors; AI as a tool |
| Does the claim match the disclosure? | Claim features that each find support in the description |
The list is TaxClue's checklist, drawn from paragraphs 5.1 to 5.2.2. The Guidelines do not prescribe formats; they ask for reproducibility.
A worked example (invented)
Nirmaan Vision files a claim for a "neural network that identifies defects" and a description that says the network may be a convolutional network, a transformer or any suitable architecture. The examiner objects to sufficiency and to abstractness under paragraph 5.2.2. Nirmaan files a response and amends to claim the specific architecture used in a steel-sheet inspection line, with the camera interface, the labelled training set and the measured defect detection result described in the original specification. The response relies on the same passages to show reproducibility and a tangible application. The claim no longer covers every possible network, but it is enabled.
Common lapses
- Listing many architectures "as options" without teaching how to implement any of them.
- Describing a result the model "may achieve" without data or an example.
- Naming an AI tool as an inventor on the application.
- Claiming a use case in one line when the description carries only the problem statement.
Need help with an AI or machine learning patent filing?
A specification for an AI invention has to teach the reader how to build and use it, and the claims must follow what is taught. Our team can review a draft against paragraphs 5.1 to 5.2.2 and help you prepare the inventor details. See patent drafting and filing and read paragraph 4.4 on sufficiency and claims alongside this article.
Key takeaways
- An abstract AI, ML or DL idea is not patentable; a tangible application may be.
- An invention created autonomously by AI is not patentable, as AI is not a person under section 6.
- An AI-assisted invention can be patented if the usual criteria and a technical effect are shown.
- The disclosure must allow reproduction without undue experimentation.
- Describe architecture, data and results, not just possibilities.
Read next
- CRI Guidelines 2025, paragraph 4.4: sufficiency, claims and means plus function
- CRI Guidelines 2025, paragraph 5.2.3: illustrative scenarios for AI claims
- Section 6 of the Patents Act, 1970: persons entitled to apply
- Guidelines on the use of AI in patent examination: scope, terms and risks
Disclaimer: Based on the manuals and guidelines published by the Office of the Controller General of Patents, Designs and Trade Marks that are named in the article, as consulted on 4 October 2026. They are guidance and do not have the force of law; the Patents Act, 1970 and the Patents Rules, 2003 as amended (including the 2024 amendment rules) prevail, and the current versions on ipindia.gov.in should be checked. This article is general information, not legal advice; check the official text before acting.