How to Patent AI & Machine‑learning Inventions in India, Practical Guidance After the 2025–26 Rule Changes

Last updated: 15 May 2026

Patenting AI inventions in India has entered a new phase. The Indian Patent Office’s revised Computer Related Invention (CRI) guidelines issued in 2025, sweeping amendments to the patent rules 2026 India framework, including overhauled Form 27 working‑statement obligations, and procedural PCT reforms 2026 now in force have collectively redrawn the landscape for anyone filing or prosecuting AI and machine‑learning patent applications in the country. For in‑house counsel, R&D heads, patent agents, and startup founders, the practical question is no longer whether AI‑related inventions can be protected in India but how to draft, file, and maintain those rights under the current rules without tripping over Section 3(k) of the Patents Act, 1970.

Executive Summary, Three Actions to Take Now

Before diving into the detail, practitioners should prioritise three immediate actions that reflect the 2025–26 changes:

• Audit Form 27 obligations. Under the amended patent rules 2026 India, every patentee (and, where applicable, licensee) must file a working statement on Form 27 within the prescribed annual window. Failure to comply can feed into compulsory‑licensing inquiries and attract penalties, a risk that is especially acute for AI products delivered as cloud services.
• Classify the invention correctly. The IPO’s 2025 CRI guidance draws a clear doctrinal line: AI‑assisted inventions, where a human contributes to conception and the AI tool augments that process, remain eligible for patent protection, provided they satisfy the technical‑effect test. Autonomously AI‑generated outputs, by contrast, are treated as non‑patentable subject matter.
• Re‑evaluate your PCT‑first vs India‑first filing strategy. The PCT reforms 2026 alter international search timelines and fee structures. Early indications suggest that for applicants targeting only a handful of jurisdictions, an India‑first national filing followed by a Paris Convention route may now be more cost‑effective than the PCT pathway in certain scenarios.

Quick Regulatory Timeline: Key Dates and What Changed

The regulatory shifts affecting AI patentability India applications arrived in quick succession. The timeline below captures the dates that matter most, together with the immediate action each requires.

Timeline of Rule Changes

Practitioners should diarise the 30 September 2026 Form 27 deadline immediately and work backward, allowing at least six weeks to collate revenue data, licensee disclosures, and manufacturing or deployment evidence, particularly where the patented AI system is delivered as a service rather than a physical product.

AI Patentability India: AI‑Assisted vs AI‑Generated Inventions

Are AI or machine‑generated inventions patentable in India? The short answer under the Indian Patent Office guidelines on AI is: inventions that are merely generated by an AI system without meaningful human intellectual contribution are not patentable, but inventions where AI serves as a tool in the hands of a human inventor can qualify, provided the standard patentability criteria and the technical‑effect test are met.

Under the Patents Act, 1970, every invention must satisfy three threshold requirements: novelty, an inventive step, and industrial applicability. For AI/ML inventions, the critical gatekeeper is Section 3(k), which excludes “a mathematical or business method or a computer programme per se or algorithms” from patentability. The 2025 CRI guidance, building on earlier IPO examination manuals and Delhi High Court jurisprudence on the “meaningful technical effect” doctrine, clarifies that the phrase per se is not a blanket bar on software‑implemented inventions. Instead, examiners are directed to ask whether the claimed invention produces a technical effect that goes beyond the normal interaction between software and hardware.

How Examiners Apply Section 3(k) to AI/ML Inventions

In practice, Indian patent examiners evaluate AI/ML claims against a two‑part test:

1. Technical contribution analysis. Does the claim, taken as a whole, solve a technical problem or produce a technical effect, for example, reduced latency in a network router, improved accuracy in a medical‑diagnostic imaging pipeline, or more efficient resource allocation in a manufacturing process?
2. Substance over form. Is the AI/ML component integrated with hardware, sensors, data streams, or a physical process in a way that goes beyond a generic computing environment running an abstract algorithm? Claims that recite a “processor executing a neural network” without tying the output to a concrete technical improvement will typically attract a Section 3(k) objection.

Industry observers expect that the Indian Patent Office will continue to narrow the scope of Section 3(k) exclusions for AI inventions that demonstrably improve a technical process, consistent with the trajectory visible in recent Delhi High Court orders applying the meaningful‑technical‑effect standard.

Inventorship and Ownership, Can AI Be Named as an Inventor?

Indian patent practice requires that at least one natural person be named as the inventor. As of May 2026, no Indian court or the IPO has accepted an AI system as a valid inventor. Applicants should name the human researchers who conceived the inventive concept and document the AI system’s contribution in the specification, for example, by describing the training methodology and the human decisions that shaped model architecture and output selection. This approach satisfies the inventorship requirement while preserving an accurate prosecution record.

Form 27 Working Statement, Who Must File, Deadlines, and Sample Template

The Form 27 working statement is the annual disclosure every patentee (and, in certain circumstances, every licensee) must file with the Indian Patent Office to demonstrate whether a granted patent is being commercially worked in India. The patent rules 2026 India amendments have tightened the reporting framework: the prescribed data fields now require more granular commercial information, and the consequences of non‑filing have been explicitly linked to compulsory‑licensing risk assessments.

Reporting Obligations by Entity Type

Special Considerations for AI/ML Products

Patenting AI inventions India through to grant is only half the compliance picture. AI products are frequently delivered as cloud‑hosted SaaS platforms rather than tangible goods, which creates practical difficulties when completing Form 27 fields that contemplate physical manufacture. Practitioners should:

• Characterise commercial working broadly. Cloud deployment, API calls processed in India, and subscription revenue from Indian customers can all constitute “working” of the patent, document these metrics carefully.
• Prepare a non‑working justification if the product is still in development. If the AI system covered by the patent has not yet been commercially deployed, include a brief narrative explaining the stage of development and anticipated launch timeline.
• Retain supporting evidence. Server logs, billing records, and customer‑deployment reports provide the evidentiary backbone for a credible Form 27 filing. Late or unsupported filings may be scrutinised during compulsory‑licensing proceedings.

Sample Form 27 Wording for an AI/ML Patent

The following illustrative language can be adapted for a granted Indian patent covering an AI‑driven medical‑imaging diagnostic tool:

“The patented invention is commercially worked in India. The AI‑based diagnostic system is deployed via cloud infrastructure and accessed by licensed healthcare facilities across India. During the reporting period, approximately [X] diagnostic analyses were processed for Indian end‑users, generating revenue of approximately INR [Y]. The patentee has not granted any licences to third parties during this period.”

Missing the Form 27 deadline can attract penalties and, critically, may be cited by third parties seeking a compulsory licence under Chapter XVI of the Patents Act on the ground that the patent is not being adequately worked in India.

Drafting & Prosecution Playbook for Patenting AI Inventions India

Drafting AI patent claims that survive Indian examination requires a deliberate strategy: every claim must anchor the AI/ML component to a concrete technical effect, and the specification must provide enough worked examples to persuade the examiner that the invention goes beyond an algorithm per se.

Six‑Step Drafting Checklist

1. Identify the technical problem. Frame the invention as a solution to a specific, measurable technical challenge, not a business objective.
2. Tie the model to hardware or a physical process. Describe how the AI/ML model interacts with sensors, actuators, network infrastructure, or data pipelines in a way that produces a tangible technical improvement.
3. Include data‑flow steps in the claims. Recite concrete steps: acquiring input data from a defined source, pre‑processing, applying the trained model, and generating an output that feeds into a downstream technical process.
4. Provide multiple worked examples. The specification should contain at least two detailed embodiments showing how the claimed method or system operates on real‑world data to achieve the stated technical effect.
5. Draft claims in three formats. File independent claims as a method, a system, and a non‑transitory computer‑readable medium storing instructions, this maximises enforcement flexibility.
6. Anticipate Section 3(k) objections. Include a paragraph in the description explicitly articulating the technical advancement achieved by the invention compared with conventional approaches.

Claim Examples

Example 1, Medical‑Diagnostic ML Model (Method Claim): “A computer‑implemented method for detecting anomalies in medical images, comprising: receiving a digital radiographic image from an imaging device; pre‑processing the image to normalise pixel intensity values; applying a trained convolutional neural network to classify regions of interest; and generating a diagnostic report with annotated probability scores, wherein the method reduces false‑positive detection rates by at least 15 % compared with conventional threshold‑based analysis.” This claim ties the AI model to a specific imaging device, recites concrete data‑processing steps, and quantifies the technical improvement.

Example 2, Recommendation Engine (System Claim): “A system for real‑time network traffic optimisation, comprising: a plurality of edge routers configured to collect packet‑flow telemetry; a processing server executing a reinforcement‑learning model trained on historical traffic patterns; and a routing controller that dynamically adjusts packet forwarding rules based on the model’s output, thereby reducing average network latency by a measurable margin.” This system claim integrates the ML model with physical network hardware and a quantifiable performance gain.

Example 3, Data‑Preprocessing Pipeline (Computer‑Readable Medium Claim): “A non‑transitory computer‑readable medium storing instructions that, when executed by a processor coupled to an industrial sensor array, cause the processor to: collect vibration data from the sensor array; apply a feature‑extraction algorithm to identify frequency‑domain signatures; input the extracted features into a trained gradient‑boosted decision‑tree model; and output a predictive maintenance alert when the model’s confidence score exceeds a predefined threshold, wherein the instructions reduce unplanned equipment downtime.” This format anchors the AI component to industrial IoT hardware and a concrete operational benefit.

Responding to Examiner Objections, Office‑Action Playbook

When the examiner issues a First Examination Report (FER) raising a Section 3(k) objection, practitioners engaged in patent prosecution India should structure the response as follows:

• Rebut the “algorithm per se” characterisation. Point to the specific claim elements that describe interaction with hardware, sensors, or a physical process, and argue that these elements, read together, produce a technical effect beyond generic computation.
• Cite the 2025 CRI guidance. Reference the IPO’s own examination framework to show that the claimed invention falls within the category of AI‑assisted inventions eligible for protection.
• Amend claims narrowly if needed. Add dependent claims that further specify the technical integration, for example, naming the type of sensor or quantifying the improvement, to give the examiner a clear allowable fallback without surrendering broad scope.
• Provide comparative performance data. Include test results or benchmarking data in the response showing that the claimed AI method outperforms prior‑art conventional approaches on a technical metric.

PCT Reforms 2026: Tactical Impact on Patenting AI Inventions India

How do the 2026 PCT reforms affect India‑origin filings and international patent strategy? In summary, the procedural changes to international search report timelines, fee payment procedures, and receiving‑office workflows mean that the cost and timing calculus for PCT‑first filings from India has shifted.

Pre‑ vs Post‑Reform Comparison

Patent teams should model both routes, PCT‑first and India‑first plus Paris Convention, using current fee data from the WIPO PCT Applicant’s Guide and the IPO’s fee schedule. For AI startups with limited budgets and a primary interest in the Indian and US markets, the India‑first route may now offer meaningful savings under the reformed PCT fee structure, while preserving the 12‑month priority‑claim window for subsequent foreign filings.

Patent Strategy for Startups: Budgets, Timelines, and Checklists

What should startups do first, file in India, file PCT, or file abroad? The answer depends on three variables: where customers and investors are located, available budget, and the maturity of the AI product.

• Step 1, File a complete or provisional application in India. This secures a priority date at a relatively low cost. Indian filing fees for startups recognised under the DPIIT Startup India programme attract significant fee concessions.
• Step 2, Within 12 months, decide between PCT and direct national filings. If the startup’s target markets include three or more countries outside India, a PCT application remains a practical vehicle despite the 2026 fee changes. For a narrower geographic focus (e.g., India plus the US), direct national filings under the Paris Convention may be more cost‑effective.
• Step 3, Budget realistically. Indian national filings (inclusive of agent fees and prosecution) typically cost significantly less than US or European filings. A PCT application entering multiple national phases can cost several multiples of the India‑only filing. Startups should reserve budget for at least one round of office‑action responses in each jurisdiction.

Where the AI model is still evolving, consider whether a provisional application, which defers full specification requirements by 12 months, is preferable to an immediate complete filing. This buys development time while locking in a priority date.

Enforcement and Licensing Considerations for AI Inventions

Securing a patent grant is the starting point; enforcing it against infringers of AI/ML systems presents distinct challenges. Unlike a physical product that can be reverse‑engineered, an infringing machine‑learning model may be deployed behind an API or embedded in proprietary cloud infrastructure, making detection difficult. Practitioners should consider the following measures:

• Draft claims with enforcement in mind. Method claims should describe externally observable outputs or measurable performance characteristics that can be verified without access to the infringer’s source code.
• Include licensing provisions for datasets and model weights. Licence agreements covering AI patents should explicitly address rights to training data, pre‑trained model weights, and derivative models to prevent gaps in protection.
• Account for data‑protection interplay. Where the patented AI system processes personal data, licensing and enforcement strategies must be consistent with India’s data‑protection framework to avoid regulatory conflict.

Practical Annexes and Templates

Practitioners navigating patent prosecution India for AI/ML applications will benefit from the following resources, which can be used alongside this guide:

• Form 27 sample template, editable working‑statement wording for AI/SaaS products, including non‑working justification language (Form 27 compliance checklist, forthcoming)
• Claim‑drafting templates, method, system, and computer‑readable‑medium formats pre‑structured for common AI use cases (drafting AI/ML patents for India, forthcoming)
• Office‑action response templates, sample rebuttals to Section 3(k) objections with CRI guidance citations (responding to IPO objections for AI inventions, forthcoming)
• PCT decision checklist, cost‑comparison worksheet for India‑first vs PCT‑first filing routes under the 2026 reforms (PCT reforms 2026 India tactical playbook, forthcoming)
• International IP protection guide, broader strategies for protecting intellectual property across borders

Conclusion

The 2025–26 rule changes have made patenting AI inventions India both more structured and more demanding. The IPO’s CRI guidance provides a clearer doctrinal framework for AI patentability, but that clarity raises the bar for claim drafting and prosecution strategy. Meanwhile, updated Form 27 obligations require ongoing compliance discipline, and the PCT reforms 2026 necessitate a fresh look at international filing economics. Practitioners who invest the time to align their applications with these changes, anchoring every claim to a demonstrable technical effect, filing working statements on time, and choosing the right cross‑border route, will be best positioned to secure and maintain meaningful patent protection for AI and machine‑learning innovations in India.

For jurisdiction‑specific guidance, consult experienced intellectual property practitioners or browse the Global Law Experts lawyer directory to connect with patent specialists in India.

Sources

1. Indian Patent Office, Official Site (Patent Rules & Notifications)
2. WIPO, PCT Applicant’s Guide (India Section)
3. Chadha & Chadha IP, Form 27 Working Statements Compliance
4. Fox Mandal, Patenting AI in India: IPO Revises CRI Guidelines
5. Lexology, AI Inventorship and Delhi High Court Orders
6. Global Patent Filing, 2026 PCT Reforms Analysis