Protecting AI & Machine Learning Innovations in India

The Section 3(k) Challenge

The Patents Act, 1970 grants protection to inventions that are novel, involve an inventive step and are industrially applicable. But it also draws boundaries to keep abstract ideas out of the patent system.

Section 3(k) · The exclusion

Mathematical methods, business methods, computer programs "per se" and algorithms are not patentable inventions.

Since AI and ML are fundamentally algorithm-driven, this clause is the central concern. The crucial word, though, is per se. The Indian Patent Office — guided by the Guidelines for Examination of Computer-Related Inventions (CRIs) — has read the exclusion to cover standalone software, not inventions that produce a genuine technical effect or contribution.

Why AI/ML isn't just "Software"

01 Learns from Data

It improves over time from data, without explicit reprogramming by a human.

02 Decides Autonomously

It classifies, predicts and makes decisions — sometimes beyond human capability.

03 Evolves Dynamically

Its adaptive, evolving nature resists a clean label of "abstract idea."

That distinction is exactly what gives applicants room to argue an AI/ML invention is more than a computer program "per se."

Eligibility & Technical Effect

An AI/ML invention becomes patent-eligible when it does more than run an algorithm — when it is tied to hardware or solves a technical problem in a technical field. That "technical effect" is the heart of the case for grant.

What Counts as a Technical Effect

Improved Efficiency

Reduced system latency, faster real-time predictions, lower power consumption.

Enhanced Performance

Higher accuracy on complex tasks beyond what prior methods achieve.

Real-world Impact

Tangible industrial benefit — predictive analytics in manufacturing, for example.

Concrete examples that tend to read as technical: AI improving real-time image processing in autonomous vehicles, an ML model optimising pharmaceutical compound discovery, or a method that cuts power use in server systems. Alongside the technical effect, the invention must still show technical advancement over what exists and industrial applicability in a real commercial setting.

Novelty & Inventive Step

Prior art in AI is broad — it includes existing patents, published research, and publicly available code, so even an open-source repository on GitHub can be cited to challenge novelty. The tougher question is usually inventive step: would the invention be obvious to a person skilled in the art?

Likely patentable

Solving a Technical Problem

AI applied in a non-obvious way that overcomes a genuine technical limitation.

Likely not

Obvious Automation

Simply automating a known manual process with AI, with no unexpected technical gain.

The Patent Process in India

The procedure mirrors any Indian patent filing — with the examiner's attention on Section 3(k) at the objection stage.

1 Prior Art Search

Novelty check

Search global databases — and code repositories — to confirm the invention is new.

2 Draft the Application

Frame the technical contribution

Articulate the technical problem and the novel solution, with diagrams and data-flow detail.

3 File with the IPO

Forms & fees

Submit the application with the required forms and fees to the Indian Patent Office.

4 Publication

18 months

The application is published after 18 months from priority, unless early publication is sought.

5 Request Examination

Within 48 months

File the examination request; the IPO assesses the application for patentability.

6 Respond to Objections

Address Section 3(k)

Reply to the examiner's report — most critically, the arguments on the 3(k) exclusion.

7 Grant of Patent

20-year term

On meeting all requirements, the patent is granted and published — protection for 20 years from filing.

Pitfalls & Best Practices

Common pitfalls

• Over-reliance on vague algorithm descriptions
• No demonstration of real-world applicability
• Broad claims that don't distinguish from prior art
• Treating the AI model as the invention, not its effect

Best practices

• Lead with the technical problem and novel solution
• Link the software to specific hardware where possible
• Use clear diagrams, workflows and data-flow explanations
• Anchor claims to a measurable technical effect

Where Grants have Happened

AI/ML patents have been granted in India where technical advancement and practical utility were clear — for example, medical-diagnostics platforms with improved accuracy, ML fraud-detection systems in banking, and smart energy-optimisation solutions combining IoT and ML.

Beyond Patents: A Layered IP Strategy

Patents are one instrument. A robust AI/ML portfolio usually combines several forms of protection.

COPYRIGHT

Code & Generated Works

Source code and certain AI-assisted works can be protected under the Copyright Act, 1957 — though only human-authored works qualify, and authorship of AI-generated output turns on human control and input.

TRADE SECRETS

Models & Datasets

Proprietary models, weights and training data are often best kept as trade secrets. India has no dedicated statute, so protection rests on contracts and common law — NDAs and access controls.

DESIGN

Interface & UI/UX

A novel graphical interface for an AI tool may qualify for protection under the Designs Act, 2000, provided it is new, original and not previously disclosed.

SEMICONDUCTOR

Chip Layouts

Layout designs of integrated circuits used in AI hardware can be protected under the Semiconductor Integrated Circuits Layout-Design Act, 2000.

Global strategy & government support

India is a signatory to the Patent Cooperation Treaty (PCT), so a single international application can reserve rights across member states — particularly valuable for startups eyeing global markets, with India as the base country.

Initiatives worth using

RESEARCH

National AI Mission

Promotes AI research and development across sectors.

STARTUPS

Startup India (SIPP)

The IP Protection scheme offers incentives and facilitator support for startups.

SUBSIDY

Patent facilitators

Provide subsidies and assistance for patent filing.

Conclusion

India's patent system is adapting to AI and ML, but the burden is on the applicant to show the invention is more than an algorithm. Understanding the Section 3(k) exclusion, demonstrating a real technical effect, avoiding the common drafting traps, and layering patents with trade secrets and design protection is what turns a breakthrough into a defensible asset.