India 2026, What the DABUS Refusal Means for Patenting AI: Drafting, Ownership and Prosecution Strategies for Businesses

The question of whether an AI inventor patent India 2026 applicants might rely upon is legally viable was answered decisively on 15 April 2026, when the Indian Patent Office refused to recognise the artificial-intelligence system DABUS as a named inventor. The decision aligns India with the majority of jurisdictions worldwide that insist on a natural person as the inventor of record, but it also raises pressing practical questions for any business whose R&D pipeline depends on machine-learning models, neural networks or autonomous design tools.

Against the backdrop of the CRI Guidelines 2025 and recent amendments to the Patents Act 1970, companies filing AI-assisted inventions in India now face a tighter regulatory environment, one that demands rigorous claim drafting, ironclad intellectual-property protection strategies, and proactive ownership documentation.

What this means for you, key takeaways:

• AI cannot be an inventor in India. Only natural persons qualify under Sections 6 and 7 of the Patents Act 1970.
• Human inventive contribution must be documented. Every filing should trace the inventive step back to an identifiable human contributor.
• Section 3(k) remains the primary obstacle. Claims that read as mere algorithms or computer programs per se will be refused; draft around the “technical effect” requirement.
• Ownership and assignment need immediate review. Employer-employee and contractor agreements must address AI-assisted outputs explicitly.
• PCT and cross-border filings should not be delayed, but descriptions and claims need tailoring for jurisdictional divergence.

1. The 15 April 2026 DABUS Refusal, Decision, Reasoning and Immediate Impact on the AI Inventor Patent India 2026 Landscape

On 15 April 2026, the Indian Patent Office published its order refusing a patent application that listed DABUS, Device for the Autonomous Bootstrapping of Unified Sentience, as the sole inventor. The application, filed by Dr Stephen Thaler, sought protection for inventions allegedly conceived autonomously by the AI system without human intervention. The refusal brings India into the growing consensus alongside the United States, the United Kingdom, the European Patent Office and Australia, all of which have rejected non-human inventorship claims in parallel DABUS proceedings.

The Office’s Reasoning: Statutory Foundations and Legal Personhood

The Indian Patent Office grounded its refusal primarily in Sections 6 and 7 of the Patents Act 1970. Section 6 stipulates that an application for a patent may be made by “any person claiming to be the true and first inventor,” while Section 7 requires the applicant to furnish proof of right. The Office concluded that the term “person” as used in the Act, and interpreted through the Indian legal framework, refers exclusively to a natural or legal person. An AI system possesses neither legal standing nor the capacity to hold or transfer rights, making it ineligible to satisfy the proof-of-right requirement.

Industry observers note that the Office also referenced the broader scheme of the Act, which contemplates inventorship as a precondition for the assignment of patent rights. Since an AI cannot enter into contracts, assign rights, or be subject to obligations, naming it as inventor would create an unbridgeable gap in the chain of title. The international intellectual property framework surrounding this question continues to evolve, but the direction of travel is clear.

Evidence and Proof-of-Right Relied Upon

A critical aspect of the refusal was the applicant’s inability to provide a conventional declaration of inventorship executed by a natural person. The Office observed that Form 1 and the accompanying declarations under the Patents Rules require a human signatory who can attest to the inventive contribution. When the applicant designated DABUS itself as the creator, no valid declaration could be furnished. The absence of a human inventor in the chain meant the application could not proceed past the formal requirements stage, regardless of the technical merits of the underlying invention.

Comparative Context: How Other Jurisdictions Have Ruled

India’s position is consistent with the majority view. The US Federal Circuit in Thaler v. Vidal upheld the USPTO’s refusal, interpreting the Patent Act’s use of “individual” to mean a natural person. The UK Supreme Court reached the same conclusion, and the EPO Boards of Appeal confirmed that an inventor designated under the EPC must be a human being. South Africa stands as a notable outlier, having initially granted a DABUS patent, though this resulted from a formalities-only examination rather than a substantive endorsement of AI inventorship. Early indications suggest that the South African approach has not been followed by any other jurisdiction and is unlikely to set a broader precedent.

2. Regulatory and Examiner Landscape: CRI Guidelines 2025 and Patents Act Amendments for Patenting AI in India

The DABUS refusal does not operate in isolation. It sits alongside the CRI Guidelines 2025, the Indian Patent Office’s updated guidance for examining computer-related inventions, and iterative amendments to the Patents Act that together define what is and is not patentable in the AI space. Understanding these instruments is essential for anyone seeking to protect intellectual property rights in AI-driven innovation.

What Examiners Now Look For Under the CRI Guidelines 2025

The CRI Guidelines 2025 instruct examiners to evaluate whether a claimed invention produces a “technical effect” or solves a “technical problem” beyond the ordinary operation of a computer. The guidelines distinguish between:

• Non-patentable subject matter under Section 3(k): A computer program per se, a mathematical method, a business method, or an algorithm presented without any technical application.
• Potentially patentable subject matter: An invention that uses a computer program or AI model as part of a system producing a demonstrable technical effect, for example, improved signal processing, reduced power consumption, faster medical-image analysis with hardware integration, or enhanced manufacturing yield.

Examiners are directed to look beyond the claim language to the substance of the specification. If the specification describes only data manipulation without linking it to a concrete technical improvement, the application will attract a Section 3(k) objection regardless of how the claims are worded.

Examiner checklist, what your application must demonstrate:

• A clearly articulated technical problem that the invention solves.
• A technical effect that goes beyond the inherent operation of a general-purpose computer.
• Integration with a physical system, sensor, actuator, network element, or industrial process (where applicable).
• Sufficient disclosure to allow a person skilled in the art to reproduce the invention, including model architectures, training parameters, and datasets (or at least representative descriptions).

3. Drafting Strategies and Claim Templates for AI-Assisted Inventions: A Practical Playbook for Drafting AI Patent Claims India

This section forms the core of this AI inventor patenting guide. After the DABUS India 2026 refusal, every patent application involving an AI or machine-learning component must be drafted with two objectives in mind: (1) survive Section 3(k) scrutiny and (2) anchor inventorship in a named natural person. The following principles and worked examples are designed to be directly usable by patent attorneys, in-house IP teams, and R&D managers.

Drafting Principle 1, Claim the Technical Effect and System Integration

The single most effective strategy for overcoming Section 3(k) AI objections is to frame the claim around the technical result produced by the invention rather than the algorithm that achieves it. The claim should describe what the system does in the physical or technical domain, not merely how data flows through a neural network.

Annotated example, Telecom ML feature:

“A method for dynamically allocating radio resources in a cellular network, the method comprising: receiving, by a base station processor, real-time channel-quality indicators from a plurality of user devices; processing the channel-quality indicators through a trained resource-allocation model to generate per-device resource-block assignments that minimise inter-cell interference; and transmitting data to the user devices using the generated resource-block assignments, wherein the method reduces average packet latency by at least 15% compared to static allocation.”

Note that the claim ties the ML model to a specific technical infrastructure (base station, radio resources, user devices), recites measurable technical improvement (15% latency reduction), and avoids presenting the model as a standalone algorithmic step.

Drafting Principle 2, Tie Claims to Physical or Technical Implementations

Where the invention involves a trained AI model, include at least one independent claim that recites a hardware or system-level element. This does not mean adding a gratuitous “processor” limitation; it means structuring the claim so that the AI component is an integral part of a technical apparatus or process that could not function without physical elements.

Annotated example, Image-processing model with sensor hardware:

“A quality-inspection system for a manufacturing line, comprising: an optical sensor array configured to capture high-resolution images of components on a conveyor; a processing unit operatively coupled to the sensor array and executing a convolutional neural network trained on a defect-classification dataset; and a reject mechanism actuated by the processing unit when the neural network classifies a component as defective with a confidence score exceeding a predetermined threshold.”

This claim integrates sensor hardware, a processing unit, and a physical reject mechanism, ensuring the AI model is embedded within a tangible technical system.

Drafting Principle 3, Data and Training Process Claims: Drafting Around Section 3(k)

Training-process claims can be patentable if they describe a novel method of preparing, curating, or transforming data that yields a measurably improved model, provided the claim is anchored to a technical application. Avoid claiming the training process in the abstract. Instead, specify the domain (medical imaging, autonomous vehicles, industrial IoT), the technical metric improved (accuracy, inference speed, false-positive rate), and any hardware-specific training steps (distributed GPU clusters, on-device fine-tuning).

Dos and don’ts for AI claim drafting in India:

• Do recite the technical problem and the measurable technical improvement.
• Do integrate claims with physical systems, sensors, actuators, or network elements.
• Do name the human inventor(s) who conceived the inventive step, document their contribution in lab notebooks.
• Don’t draft claims that read as pure mathematical formulae or flowcharts of data manipulation.
• Don’t present the AI model as a “black box” without disclosing architecture, training data characteristics, or performance benchmarks.
• Don’t conflate the AI’s output with the inventive contribution, the human must have directed, designed, or adapted the system.

Worked Examples, Three Short Sample Claim Sets

Example A, Process claim with physical step (predictive maintenance):

“A method for predictive maintenance of an industrial turbine, comprising: collecting vibration data from accelerometers mounted on turbine blades at predetermined intervals; feeding the vibration data into a recurrent neural network trained to detect anomaly signatures indicative of bearing degradation; generating a maintenance alert when the neural network outputs a degradation-probability score exceeding 0.85; and automatically adjusting turbine operating parameters to reduce rotational speed until maintenance is performed.”

Example B, System claim (autonomous logistics):

“A warehouse routing system comprising: a fleet of autonomous guided vehicles (AGVs) each equipped with LiDAR sensors; a central routing server executing a reinforcement-learning-based path-planning module; and a collision-avoidance subsystem that overrides the path-planning module when LiDAR data indicates an obstacle within a safety threshold, wherein the path-planning module dynamically re-routes AGVs to reduce average order-fulfilment time.”

Example C, Apparatus claim (medical diagnostics):

“A retinal imaging apparatus comprising: a fundus camera configured to capture retinal scans; a diagnostic processor executing a deep-learning classifier trained on a labelled dataset of diabetic-retinopathy images; and a display unit presenting a graded severity map to a clinician, wherein the classifier achieves a sensitivity of at least 95% on the labelled dataset.”

Each of these examples roots the AI component within a physical apparatus or technical process, names measurable outcomes, and leaves space for the specification to identify the human inventors who designed, trained, and validated the system.

4. Inventorship vs Ownership of AI Inventions India, Attribution, Proof of Right, and Contractual Safeguards

The DABUS refusal sharpened a distinction that many organisations have historically treated loosely: the difference between inventorship (who conceived the invention) and ownership (who holds the patent rights). After 15 April 2026, getting this distinction wrong can invalidate an entire filing.

Employer-Employee Inventions, Best-Practice Contract Clauses

Under Indian law, inventions made by employees in the course of their employment typically vest in the employer, but only if the employment agreement expressly provides for this. With AI-assisted inventions, the contract should go further:

• Define “AI-assisted invention” explicitly to include outputs generated with substantial use of employer-provided AI tools, models, or datasets.
• Require the employee to maintain contemporaneous records of their inventive contribution, distinguishing it from the AI’s computational output.
• Include a pre-assignment clause covering all inventions arising from the employee’s use of the employer’s AI infrastructure.

Contractors and Third-Party AI Providers, Assignment and Disclosure Clauses

When a company uses a third-party AI platform (cloud-based ML services, proprietary models licensed from vendors), the ownership position becomes more complex. The contract with the AI provider should address:

• Who retains IP rights in outputs generated by the provider’s model when applied to the company’s data.
• Whether the provider has any residual claim to inventions arising from model fine-tuning or custom training.
• Obligations for the provider to disclose model architecture and training-data provenance if needed for patent prosecution or litigation.

Evidence Logs and Reproducibility Records

The Indian Patent Office’s emphasis on proof of right means that companies must maintain robust documentation linking every claimed invention to a natural person’s inventive contribution. Recommended records include model-training logs, version-control histories, experiment notebooks, and internal invention-disclosure forms signed by the contributing engineers or scientists.

5. PCT and Cross-Border Prosecution Strategy for AI Filings Originating in India

The DABUS India 2026 refusal does not change the procedural mechanics of PCT filings, but it significantly affects how applicants should draft their descriptions and claims to preserve options across jurisdictions that may treat AI inventorship and patentable subject matter differently. A well-designed PCT strategy for AI filings must account for divergent examination standards from day one.

PCT Filing: Claims and Description Drafting Tips to Preserve Options

• File the Indian national application first where the invention originated in India, Section 39 of the Patents Act 1970 restricts filing abroad without prior permission if the invention was made in India. Obtain a foreign-filing licence or file the Indian application at least six weeks before the PCT filing.
• Draft the PCT description with layered disclosure: include a detailed description of the technical effect and system integration (for India and the EPO), the specific technical improvement and machine-or-transformation test analysis (for the US), and a clear human-inventorship narrative throughout.
• Use dependent claims strategically: include narrower dependent claims that emphasise hardware integration (useful for India) and broader dependent claims that focus on the AI-method steps (potentially allowable in the US or other jurisdictions with more permissive subject-matter standards).
• Preserve priority rigorously: any delay in filing can be exploited by competitors or create prior-art problems. The 12-month Paris Convention / PCT priority window remains the critical timeline.

Where to Emphasise Technical Contribution

For the Indian national phase, the specification should front-load the technical problem and technical solution sections, with the AI model described as a means of implementation rather than the core inventive concept. For the US national phase, the specification should include detailed evidence of a practical application to pre-empt Alice/Mayo rejections. For the EPO, emphasise the “further technical effect” beyond the normal operation of a computer.

Cross-border filing checklist:

• Confirm foreign-filing licence or Section 39 compliance.
• Draft a single specification that supports narrowing for India (technical effect) and broadening for the US (method claims).
• Name all human inventors with supporting declarations, do not leave any inventor field blank or designate an AI system.
• Include performance benchmarks and reproducibility data in the specification.
• Prepare assignment documents for each jurisdiction in advance of national-phase entry.

6. Litigation and Enforcement Risk Map, Likely Challenges and Defensive Documentation

As AI-assisted patents proliferate, industry observers expect three categories of challenge to emerge most frequently in Indian patent litigation and opposition proceedings:

• Inventorship attacks: Opponents may argue that the true inventor was the AI system, not the named human, thereby invalidating the patent for incorrect inventorship. Defence requires contemporaneous records of the human inventor’s design choices, model-selection decisions, and experimental validation.
• Prior art in training data: If the AI model was trained on publicly available data, an opponent could argue that the invention was already disclosed in the training corpus. Mitigation requires detailed documentation of what is novel, the system architecture, the application context, or the combination of known elements in a new technical configuration.
• Enablement and sufficiency: A vague description of the AI model (“a neural network was trained”) without architecture details, hyperparameters, and training methodology may fail the sufficiency requirement under Section 10. Provide enough detail for a skilled person to reproduce the invention.

Mitigation checklist:

• Maintain version-controlled model repositories with timestamps.
• Log all human interventions in the training, tuning, and evaluation pipeline.
• Preserve dataset provenance records (source, date acquired, any transformations applied).
• Keep signed invention-disclosure forms that pre-date the filing.

7. Practical Checklist, Pre-Filing and Prosecution Steps for AI Inventor Patent India 2026 Applications

Use this checklist as a client-ready template before filing any AI-related patent application in India:

1. Identify all natural persons who made an inventive contribution, interview each and document their specific role.
2. Review employment and contractor agreements for IP assignment clauses covering AI-assisted outputs.
3. Collect and preserve model-training logs, version-control records, and experiment notebooks.
4. Draft claims around the technical effect and system integration, not the algorithm in isolation.
5. Ensure the specification discloses model architecture, training data characteristics, and performance benchmarks.
6. Prepare Form 1 declarations with accurate inventor details, never leave the inventor field blank.
7. Confirm Section 39 compliance if a foreign or PCT filing is planned.
8. Draft the PCT specification with layered disclosure to accommodate divergent jurisdictional standards.
9. Obtain and file assignment documents before or simultaneously with the patent application.
10. Set calendar reminders for 12-month priority deadline, 31-month national-phase entry, and any foreign-filing-licence timelines.
11. Brief the prosecution team on CRI Guidelines 2025 examiner expectations and prepare responses to anticipated Section 3(k) objections.
12. Engage patent counsel experienced in AI prosecution to review the complete application before filing.

Conclusion, Act Now to Secure Your AI Patent Portfolio

The 15 April 2026 DABUS refusal is not a setback for businesses innovating with AI, it is a clarification. The rules are now unambiguous: patent protection for AI-assisted inventions in India is available, but only when a natural person is identified as the inventor and the claims demonstrate a genuine technical effect beyond mere computation. Companies that act promptly, reviewing their inventorship documentation, tightening employment and contractor agreements, and adopting the claim-drafting strategies outlined in this AI inventor patent India 2026 guide, will be well positioned to build defensible patent portfolios. Those that delay risk invalidation, ownership disputes, and lost priority. The time to review your AI patent strategy is now.

Last updated: 1 May 2026. This article will be updated when new orders or appeal decisions related to DABUS India are published.

Sources

1. Intepat, Can AI Be an Inventor in India? DABUS Refused
2. SpicyIP, The Inventor is still Human: Indian Patent Office’s DABUS Refusal
3. BananaIP Counsels, Inventorship of DABUS in India
4. Indian Patent Office, Patents Act 1970 and Official Guidance
5. Fox Mandal, CRI Guidelines 2025 Commentary
6. Chambers Practice Guides, Patent Litigation 2026 (India)
7. MLex, Stephen Thaler’s AI-Generated Patent Application Rejected by Indian Patent Office
8. IIMA, AI Inventor Debate Under Patent Law: Post-DABUS Comparative Analysis
9. NASSCOM, Patent Pulse 2025: Decoding India’s Ascent in the AI Patent Landscape