An AI or machine learning invention can be patented in India if it is applied to a technical problem and produces a technical effect: for example, detecting defects from sensor data, improving medical image analysis, or reducing the compute needed to train or run a model. Algorithms and mathematical methods by themselves are excluded under Section 3(k). The inventor must be a human: an AI system cannot be named as the inventor.
- Registered patent agentsPatent work is handled only by agents registered with the Indian Patent Office
- NDA before you shareWe sign a non-disclosure agreement before you send any technical details
- Your idea stays yoursNothing is filed or published without your written approval
- Free first reviewA registered patent agent tells you if it is worth filing, before you pay
Usually patentable
- AI applied to a technical field: image, speech or signal processing, robotics, diagnostics, predictive maintenance
- Methods that make training or inference faster, smaller or less power-hungry on real hardware
- A specific pipeline for processing sensor or machine data that gives a technical result
- Hardware designed for AI workloads
Usually not patentable
- A model, neural network architecture or training algorithm claimed in the abstract
- AI used for a business purpose only: pricing, ad targeting, credit scoring with no technical effect
- Generic 'use AI to do X' claims with no detail of how
- The dataset itself (protect as a trade secret or by contract)
Where AI fits in Indian patent law
There is no separate AI law: AI inventions are examined under the same rules as other computer-related inventions. Section 3(k) excludes "mathematical methods", "computer programmes per se" and "algorithms", and a machine learning model is, at its core, all three. So the question is always: what technical problem does the AI solve, and how?
Following Ferid Allani v Union of India (Delhi HC, 2019), examiners look for a technical effect or technical contribution. For AI, the clearest cases are where the model works on real-world technical data (images, signals, sensor readings) or improves how computing resources are used.
Examples: patentable vs excluded
| Likely excluded | Likely patentable |
|---|---|
| A new neural network for classifying data | A method of detecting cracks in welds from X-ray images using a specific pre-processing and classification pipeline |
| An ML model predicting customer churn | A model predicting bearing failure from vibration sensor data, triggering a machine shutdown |
| A chatbot that answers customer queries | A speech-recognition method that reduces errors in noisy environments using a specific signal-processing step |
| A recommendation algorithm | A method of compressing a model so it runs on a low-power device with a defined accuracy and memory limit |
How to draft an AI patent application
- Name the technical field and problem at the start: "detecting defects in castings", not "an AI platform".
- Describe the whole pipeline: data capture, pre-processing, feature extraction, model, output and what the output controls.
- Give enough detail to reproduce it (sufficiency of disclosure): the type of model, how it is trained, the kind of training data, and key parameters. A 'black box' description invites objections.
- Tie the output to a technical action: shutting a valve, flagging a scan, adjusting a motor, allocating compute.
- Show results: accuracy, speed or resource figures compared with known methods.
Who is the inventor of an AI-generated invention?
The Patents Act requires the application to name the "true and first inventor", and the law assumes this is a person. Courts and patent offices in the UK, US and Europe have refused applications naming an AI system (the DABUS cases), and Indian practice is the same. Name the people who designed the system, framed the problem and selected or verified the result.
Before you publish or open-source
Research papers, preprints, GitHub repositories and conference demos are all prior art against you. India gives only a narrow exception for certain disclosures (for example, at government-notified exhibitions or before learned societies, within 12 months). The safe course is to file a provisional application before any public disclosure.
Frequently asked questions
Can ChatGPT-style or generative AI inventions be patented in India?
The general idea of generative AI cannot. A specific technical improvement, for example a method that reduces the memory needed to run a model, or applies generation to a technical problem such as designing a component under engineering constraints, may be patentable.
Can an AI be listed as the inventor in India?
No. The inventor must be a natural person. Name the people who made the inventive contribution.
Is my training dataset protected by a patent?
No. Data is not an invention. Protect datasets with confidentiality, access controls and contracts; some compilations may also have copyright protection.
Will an AI patent application get a Section 3(k) objection?
It is common for AI applications to receive one in the First Examination Report. A specification that clearly sets out the technical problem and effect gives you strong grounds to reply.
Patent guides for other technologies
Software and apps · Pharma and formulations · Medical devices and diagnostics · Mechanical and product inventions · EV, battery and clean-tech · Agritech and food processing
Not sure about your invention? Try our free patentability checker.
Also read: What cannot be patented in India · How to patent an idea in India
This guide is general information, not legal advice. Whether your invention can be patented depends on its details and the prior art; speak to a registered patent agent about your case.