AI & Intellectual Property 2026 Comparisons

India has no AI-specific IP legislation. The intersection of AI and IP is nonetheless actively contested, and the courts are seized of it. The Copyright Act, 1957, the Patents Act, 1970, the Trade Marks Act, 1999 and the Designs Act, 2000 govern the field, and none carves out an exemption or exception for AI-generated output.

The Copyright Act, 1957 did not contemplate AI-based data processing. It does, however, contain a technology-neutral provision that predates generative AI but applies as a principal legislation that is guideline-based rather than being granular and prescriptive. The Copyright Act, 1957 treats the author of a literary, dramatic, musical or artistic work that is computer-generated as the person who causes the work to be created.

India is a party to the Berne Convention, Paris Convention, TRIPS Agreement, Patent Cooperation Treaty, Madrid Protocol, WIPO Copyright Treaty and WIPO Performances and Phonograms Treaty.

Foreign AI developers are not governed by any separate body of law, and are treated on equal footing as domestic right-holders. They stand on the same footing and receive national treatment under the Copyright Act, 1957, and no nationality-based rule applies to foreign AI developers. In a recent interim order passed in the precedent-setting matter of ANI Media Private Limited v OpenAI OpCo LLC, the Delhi High Court held prima facie that it had territorial jurisdiction over a foreign AI developer that maintained no servers in India, under Section 62(2) of the Copyright Act, 1957 and Section 20 of the Code of Civil Procedure, 1908, because the developer targeted Indian users and the plaintiff carried on business in India.

AI systems, generative AI, foundation models, general-purpose AI models, computer-generated works, AI-assisted inventions, training data, weights, prompts or outputs are not defined under any specific legislation in India.

The only adjacent statutory concept is the computer-generated work under Section 2(d)(vi) of the Copyright Act, 1957, which the statute does not itself define. Outside IP legislation, the IT Rules, 2021, as amended with effect from 20 February 2026, define “synthetically generated information”, but that definition operates for intermediary due diligence and does not allocate IP rights.

Autonomous and agentic AI systems attract no distinct treatment, and no specific IP classification applies to them. Liability follows the person who controls or authorises the system’s conduct.

No dedicated AI and IP tribunal exists in India.

Commercial courts and the commercial divisions of the high courts hear and adjudicate upon civil infringement proceedings, and several high Courts, including Delhi, Madras and Calcutta, maintain specialist intellectual property divisions. The Supreme Court is the final appellate court for all civil matters. The Controller General of Patents, Designs and Trade Marks (CGPDTM) oversees the Patent Office, the Trade Marks Registry, Designs Wing, and Geographical Indications Registry under the Department for Promotion of Industry and Internal Trade (DPIIT).

No specialist procedures or enforcement priorities apply to AI-related IP rights. The closest examination guidance is the Guidelines for Examination of Computer Related Inventions, 2025, issued by the CGPDTM on 29 July 2025, which address AI, machine learning and adjacent subject matter under Section 3(k) of the Patents Act, 1970.

Original source code, object code, documentation, architecture diagrams, interface graphics and sufficiently expressive prompt libraries can be protected under the Copyright Act, 1957.  Qualifying technical systems and processes can be protected under the Patents Act, 1970. Branding of an AI-tool can be protected under the Trade Marks Act, 1999. Qualifying visual features can be protected under the Designs Act, 2000.

Computer programs are protected as literary works against unauthorised reproduction, electronic storage, adaptation and communication of protected expression under the Copyright Act, 1957. Original architecture descriptions, system instructions, prompts and prompt libraries can qualify for protection under the Copyright Act, 1957 where the literary work, selection and arrangement reflect creativity and human authorship.

The Copyright Act, 1957 does not protect algorithms, mathematical relationships, methods, functionality, API or desired model outcomes. The Act, does not specifically exclude model weights and intermediate computational states, but to ensure protection under it, such model weights and intermediate computational states must have identifiable original expression and human authorship. There is no protection under the Copyright Act, 1957 that covers algorithms, mathematical relationships, methods, functionality, APIs or desired model outcomes. Indian copyright laws do not expressly exclude model weights or intermediate computational states, but the level of protection depends on identifiable original expression and human authorship, which such artefacts will rarely exhibit in a manner that can be clearly defined, delineated and demonstrated from an evidentiary standpoint.

AI-related inventions are assessed under the Patents Act, 1970, with the ordinary requirements of novelty, inventive step, industrial applicability and sufficient disclosure. The Patents Act, 1970 excludes mathematical methods, business methods, computer programs per se and algorithms. The “per se” qualifier is in relation to computer programs, but business methods are also completely excluded, without any technical-effect analysis.

Ferid Allani v Union of India, Microsoft Technology Licensing LLC v Assistant Controller of Patents and Designs and Raytheon Company v Controller General of Patents and Designs, read with the Guidelines for Examination of Computer Related Inventions, 2025 (the “CRI Guidelines”) issued on 29 July 2025, establish that although computer programs per se are not patentable, a computer-related invention that demonstrates a tangible technical contribution or technical effect beyond ordinary digital execution may be patented. The assessment however would be made on a case-by-case basis, and the substantive patentability requirements do not differ from those applying to usual protection for patentable inventions.

Models, weights, datasets, prompts, system instructions, evaluations, agent permissions and deployment know-how are protected primarily through contractual confidentiality clauses.

Reasonable measures include execution of non-disclosure agreements, least-privilege access, encryption, logging and internal data-loss prevention measures. Public information and independently developed information are not protected under IP legislation.

Disclosure to a court or regulatory authority will not, in itself, constitute a waiver or loss of confidentiality, provided that the disclosure is made pursuant to appropriate safeguards, including obtaining a protective order, or other restricted-access procedures.

As a stark contrast, publishing a model card or releasing model weights on an open-access basis erodes confidentiality, but only to the extent of the information thereby made public.

The position does not differ from the usual protection of confidential information. India has no standalone trade secrets statute, and protection rests on contract, equity and the common law duty of confidence.

Datasets and databases receive copyright protection only where their selection or arrangement is original. Original taxonomies, annotation manuals, benchmark questions and synthetic examples can qualify separately or as parts of a compilation under the Copyright Act, 1957.

Non-original collections rely principally on contractual confidentiality clauses, API terms and internal technical measures, supported by Sections 43 and 66 of the Information Technology Act, 2000, which address unauthorised access to a computer resource. India recognises no sui generis database right.

Downloading, scraping, storing and pre-processing copyrighted works for AI training may amount to copyright infringement. The courts determine the question case by case.

In ANI Media Private Limited v OpenAI OpCo LLC, the Delhi High Court held on 24 July 2026, on a prima facie basis, that OpenAI’s storage and use of lawfully accessed news articles to train its large language models fell within fair dealing for research under Section 52(1)(a) of the Copyright Act, 1957. The Delhi High Court reasoned that the use was non-substitutive and that ANI Media had not demonstrated market harm. That finding was made on an interim application; the suit remains pending and the order is open to appeal before a Division Bench.

Fine-tuning, retrieval-augmented generation and tool-mediated access carry materially higher risk where they reproduce source passages in a manner approaching transcription, or where they support paywall circumvention. Lifting text wholesale, without analysis, is the highest-risk class of data acquisition by AI systems in India.

There is no express text-and-data-mining exception in India. Section 52 of the Copyright Act, 1957 provides defences for fair dealing for private or personal use, including research, and for criticism or review, the reporting of current events, and specified educational, library and archival uses. In ANI Media Private Limited v OpenAI OpCo LLC, the Delhi High Court held prima facie that the commercial character of the user does not, by itself, defeat the research limb of fair dealing. Market harm nonetheless remains material, and the High Court gave weight to the absence of evidence of lost subscribers, reduced licensing revenue, traffic diversion or market substitution.

Unauthorised access, misuse of credentials, paywall circumvention or reliance on pirated sources will ordinarily defeat the fair-dealing defence and can attract offences under the Information Technology Act, 2000.

Areas of certainty are as follows.

• Storage or copying of publicly accessible content for LLM training can qualify as fair dealing under the Copyright Act, 1957 even where the resulting product is commercial, at least at the prima facie stage.
• Indian courts can assert jurisdiction over foreign AI developers where the plaintiff resides or carries on business in India and the developer targets Indian users.
• Retrieval-augmented output that synthesises a copyrighted work without reproducing its expression verbatim is unlikely to be treated as infringing.

Areas of uncertainty are as follows.

• ANI’s claims of memorisation and regurgitation, and its claims arising from hallucinated attributions, remain pending on the merits before the Delhi High Court.
• Whether the Delhi High Court’s findings on the interim application can be extended to non-text or non-news content such as music, images and books.

The Delhi High Court’s findings arose on an interim application for a temporary injunction and do not finally determine the suit. The Delhi High Court expressly confined its observations to that application. The proceedings remain pending, the order is amenable to appeal before a Division Bench, and the judgment on the merits should resolve much of the present uncertainty.

India does not have a settled market practice for voluntary licensing of copyright works for AI-training purposes.

India has established copyright societies, including the Indian Performing Right Society Limited (IPRS), Phonographic Performance Limited (PPL) and the Indian Singers’ Rights Association (ISAMRA), which administer copyright and related rights and collect royalties under the Copyright Act, 1957. The tariff schemes administered by such societies are required to be fair, reasonable and non-discriminatory. The DPIIT-appointed Committee on Generative Artificial Intelligence and Copyright has, however, rejected both a zero-price text-and-data-mining exception and a purely voluntary licensing market as the appropriate framework for AI-training data.

The DPIIT Working Paper on Generative AI and Copyright (Part I), titled “One Nation, One License, One Payment: Balancing AI Innovation and Copyright”, was released in December 2025 for public consultation, with comments initially invited by 7 January 2026 and subsequently extended, with further submissions accepted until 6 February 2026. The Working Paper proposes statutory blanket licensing, without a rights-holder opt-out, to be administered by a newly constituted collective body known as the Copyright Royalties Collective for AI Training (CRCAT). Under the Working Paper, AI developers would be permitted to train AI systems on lawfully accessed content without negotiating individual licences, with royalties becoming payable on commercialisation of the relevant AI system. Rates would be fixed by a government-appointed rate-setting committee and would be subject to judicial review, and developers would file an AI training data disclosure form with CRCAT.

India currently has no mechanism that specifically addresses orphan works or out-of-commerce works in the context of AI training. Additionally, India has no settled market rate or customary remuneration structure for AI-training licences. The Indian market for remuneration relating to training data therefore remains nascent and unsettled, pending judicial clarification in the proceedings in ANI Media Private Limited v OpenAI OpCo LLC.

India does not have a statutory text-and-data mining opt-out mechanism. Technical and contractual measures such as robots.txt files, “no-AI” metadata, content credentials, website terms of use and licensing registers do not, in themselves, create or confer exclusive rights.

In ANI Media Private Limited v OpenAI OpCo LLC, the availability to ANI of crawler-blocking measures weighed in the Delhi High Court’s refusal of interim relief. The decision did not hold that copyright protection is contingent upon a rights-holder having opted out of AI-related scraping or data mining. Additionally, an autonomous agent’s failure to comply with a notice would not, by itself, absolve its operator of responsibility. The circumvention of technological protection measures and the removal or alteration of rights-management information independently attract liability under Sections 65A and 65B of the Copyright Act, 1957.

No general Indian IP rule requires publication of a complete training-data inventory. However, AI developers must retain records of:

• methods and sources of data acquisition;
• licences and reservations of rights;
• filtering and deduplication methodologies;
• responses to removal requests;
• dataset version histories; and
• use of synthetic data.

The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, as amended with effect from 20 February 2026, require covered intermediaries to label specified synthetically generated information and to embed persistent provenance metadata. Those obligations are directed at content governance and neither create nor allocate IP rights. Beyond that defined scope, broader AI governance guidance in India remains largely non-binding.

In cross-border matters, liability is determined on a case-by-case basis and by identifying whether copying, storage, communication, or authorisation occurred, and where the relevant services or outputs were targeted. In ANI Media Private Limited v OpenAI OpCo LLC, the Delhi High Court found, prima facie, a sufficient Indian territorial connection notwithstanding the use of overseas training infrastructure, treating storage abroad as the terminal step in a chain that began with access to and transmission of the works from India.

A claimant must establish:

• subsistence and ownership of the relevant IP right;
• performance in India of a restricted act, such as copying, reproduction, adaptation, extraction, communication to the public, distribution or commercial use;
• use of the whole or a substantial part of the protected subject matter; and
• absence of a licence, consent or statutory exception.

Conduct undertaken automatically by an AI agent or tool-using system is attributed to the provider, operator or person who authorised or controlled it.

In ANI Media Private Limited v OpenAI OpCo LLC, the Delhi High Court held, at the interim stage, that OpenAI had stored ANI’s copyrighted works, that the storage was prima facie protected by the research limb of fair dealing, and that the challenged outputs were not substantially similar to ANI’s original expression. The findings of the Delhi High Court are interim. The precise application of infringement principles to AI training and deployment remains unsettled in pending proceedings.

Indian courts have not yet determined that model weights, parameters or embeddings constitute infringing copies. The principal issue is whether the artefact fixes or embodies protected expression, or a substantial part thereof, in a form capable of reproduction, extraction or communication.

However, caches, retrieval indexes and memorised sequences that can predictably reproduce protected passages present a high risk of constituting infringement. The mere fact that an artefact was derived from training data would not, by itself, establish infringement. In ANI Media Private Limited v OpenAI OpCo LLC, the Delhi High Court found no prima facie evidence of memorisation, regurgitation or substantial similarity on the material before it. The court’s findings do not establish a general immunity for model weights or other internal model artefacts.

A model provider is not automatically liable for every infringing output generated by a user or downstream system. Liability arises where the provider authorises, procures, induces or materially facilitates the infringement, exercises control over the relevant tools or publication process, has specific knowledge of infringement, or continues the conduct after notice.

Section 79 of the Information Technology Act, 2000 confers safe harbour on an intermediary that observes the prescribed due diligence requirements, including the expeditious removal of infringing content on obtaining actual knowledge. Whether a generative AI provider qualifies as an intermediary is itself contested. In IndiaMART InterMesh Limited v OpenAI Inc., decided on 20 May 2026, the Calcutta High Court observed prima facie that ChatGPT is better characterised as an originator than an intermediary, while describing the question as complicated and vexed and leaving it for final determination.

The risk of liability for a model provider is higher where the autonomous agent:

• has broad and undefined access permissions;
• is designed to perform functions that can lead to infringement;
• hosts, uploads and transmits information without a human-in-the-loop; or
• continues operating despite receiving notice or becoming aware of such infringement. The risk of liability for the model provider is lower where the autonomous agent has restricted access permissions, requires a human-in-the-loop before publication, applies effective safeguards and maintains appropriate monitoring and takedown procedures.

Liability arises where confidential information is acquired, accessed, retained or used in breach of a non-disclosure agreement, employment obligation, restricted API term, data licence or other duty of confidence.

Unauthorised use of such information for training, fine-tuning, prompting, model-memory storage, tool-assisted retrieval or evaluation constitutes misuse, even where the information has not been publicly disclosed.

The claimant must identify the information with sufficient particularity, establish its confidential nature, prove that the defendant obtained or accessed it, and demonstrate unauthorised use, disclosure, retention or threatened disclosure.

Potential defences include prior public disclosure, independent development and authorised use. Available remedies may include injunctions, preservation orders, access restrictions, segregation or deletion of the information from relevant systems, damages and other appropriate relief.

An AI developer or model provider can raise both substantive and procedural defences. The principal substantive defence is fair dealing under Section 52 of the Copyright Act, 1957, and in particular Section 52(1)(a), on which the Delhi High Court relied at the interim stage in ANI Media Private Limited v OpenAI OpCo LLC.

Other substantive defences include the absence of a substantial taking, the idea-expression and fact-expression distinctions, independent creation, an express or implied licence arising from platform terms, robots directives or open-source and Creative Commons terms, and exhaustion in relation to lawfully acquired copies. For patents, the research and experimental use exemptions under Sections 47 and 107A of the Patents Act, 1970 are available. For trade mark claims, honest and descriptive referential use under Section 30 of the Trade Marks Act, 1999, and the absence of use in the course of trade, are the principal answers.

Procedurally, a defendant can rely on limitation, being three years from the accrual of the cause of action under the Limitation Act, 1963, with a fresh period accruing on each continuing act of infringement; on delay, acquiescence and estoppel in resisting interim relief; on the absence of a prima facie case, an adverse balance of convenience or the adequacy of damages; and on defects in territorial jurisdiction or in the joinder of parties. Section 79 of the Information Technology Act, 2000 offers a safe harbour where the provider qualifies as an intermediary, although IndiaMART InterMesh Limited v OpenAI Inc. casts prima facie doubt on that characterisation for generative systems.

Public interest is not a freestanding statutory defence, but the Delhi High Court in ANI Media treated the public benefit associated with the development of large language models as material to the balance of convenience. Competition-law arguments, including allegations of abuse of dominance in the licensing of training content, are pursued before the Competition Commission of India rather than raised as a defence in an infringement suit.

An AI-generated output infringes copyright where it reproduces, adapts or communicates the whole or a substantial part of protected expression without authorisation. The risk of infringement is higher where the output contains verbatim text, source code, music, images or other distinctive expression from a training work, prompt, reference image or retrieval-augmented source. A prompt, reference image or retrieved source supplies direct evidence of the causal connection between the source material and the generated output.

Copyright does not protect facts, ideas, themes, genres, styles or techniques. Accordingly, imitation of an artist’s style, genre, voice or technique will not constitute copyright infringement unless protected expression from an identifiable work has also been reproduced. Misleading attribution or an implied association gives rise to the risk of passing-off, personality-rights, moral-rights or consumer-protection claims.

A user, deployer or customer can infringe IP by generating, publishing, selling, importing or otherwise exploiting an infringing output, regardless of intent. Intention bears on remedies, including punitive damages and costs, rather than on the existence of infringement. The risk is aggravated where the user, customer or deployer supplies protected material, circumvents safeguards, repeatedly extracts protected content, or commercialises the output without adequate human review. Generic prompts, human review, and reliance on the model provider’s terms and conditions reduce the risk of liability to the user, but do not preclude liability for infringement. Provider warranties or indemnities generally allocate risk between the contracting parties and do not affect the rights of third-party rights-holders.

An AI-generated output infringes a registered trade mark where an identical or deceptively similar mark is used in the course of trade in relation to the relevant goods or services.

Passing off is a common law action, preserved by Section 27(2) of the Trade Marks Act, 1999, and requires proof of goodwill, misrepresentation and likely damage.

The risk of infringement is aggravated where an AI-generated logo, product image, advertisement, packaging, domain name, or other material:

• reproduces a third party’s mark or logo;
• imitates distinctive product get-up;
• suggests sponsorship, approval or affiliation;
• uses a celebrity’s name, image, voice or other personality indicia; or
• creates a false impression of origin or endorsement.

A “brand style” is not independently protected under the Trade Marks Act, 1999. However, distinctive trade dress, characters, logos, slogans and other source-identifying features are protected under the Trade Marks Act, 1999.

An AI-generated technical description or instruction does not, by itself, constitute patent infringement merely because it describes a patented invention. Infringement arises where a person, in India, makes, uses, offers for sale, sells or imports a patented product, or uses a patented process, in contravention of the Patents Act, 1970.

AI-generated software can infringe copyright and also facilitate the unauthorised operation, manufacture or use of a patented system. Under Section 22 of the Designs Act, 2000, the unauthorised application, importation or sale of an obvious or fraudulent imitation of a registered design constitutes design piracy.

Indian law does not recognise a single doctrine of indirect patent infringement. However, intentional procurement, authorisation, inducement or participation in a common design can give rise to fact-specific liability under applicable statutory, contractual or common-law principles.

Indian law attributes the conduct of an AI agent to the responsible natural or legal person, and does not recognise the autonomous agent as an independent legal actor. Each act (including scraping, API access, retrieval, code generation, uploading, publication and product or design selection) is assessed by reference to the person’s authority, control, knowledge and the territorial nexus of the conduct.

The risk of liability is aggravated where a business determines the autonomous agent’s objectives, provides access credentials, grants extensive permissions or enables automated publication without human review.

Copyright protection in AI-assisted works requires identifiable original expressive choices made by a human. The strength of the claim increases where a human exercises detailed direction, iterative control, editing, rewriting, selection, arrangement or combination of these elements.

A detailed prompt may itself qualify for copyright protection, but that protection does not extend automatically to the AI-generated output. Selection and arrangement can support compilation copyright even where individual outputs are unprotected.

Section 2(d)(vi) of the Copyright Act, 1957 separately attributes a computer-generated work to the person who causes it to be created. How that provision interacts with the requirements of originality and human creativity remains unsettled.

In agentic systems, prompt chaining, prompt decomposition, meta-prompting, autonomous task planning, and automated selection of intermediate outputs weaken the claim to human authorship unless followed by meaningful human creative curation, direction and control.

India recognises no separate sui generis right in AI-generated output. Section 2(d)(vi) of the Copyright Act, 1957 attributes a computer-generated literary, dramatic, musical or artistic work to the person who causes it to be created. However, no binding Indian judicial precedent has clarified the application of Section 2(d)(vi) to modern generative-AI systems.

Depending on the facts of the claim, such a person may be the user, developer or deployer, having regard to causation and creative responsibility. An AI system is not a legal person and cannot itself own copyright. The position remains unsettled on the requirements of originality, first ownership, moral rights and the relationship between Section 2(d)(vi) and the requirement of human expression for literary works.

The primary requirement for joint authorship is collaboration between human authors whose contributions are inseparable or indistinguishable in the resulting output. An AI system cannot presently hold copyright or an ownership interest. Accordingly, a provider and a user do not become joint authors merely because the provider supplies the technology and the user supplies prompts.

An output derived from an existing work may constitute an adaptation or reproduction where it incorporates the whole or a substantial part of the protected expression. Such use requires authorisation unless an applicable statutory exception (such as fair dealing) applies. A human adapter acquires rights only in the original elements contributed by such person and cannot exploit the underlying work beyond the scope of the relevant licence. Human editing, selection or arrangement can attract separate copyright protection but does not cure infringement in any unauthorised copied material.

Ownership depends on the assignment, licensing, commissioning, and contractual or collaborative arrangements.

Copyright subsists in India without a registration requirement. Under the Copyright Act, 1957 there is no general obligation to disclose the use of AI. However, applicants must be able to accurately identify the work, author and claimant. Representing substantially autonomous material as wholly human-authored can lead to refusal of registration, rectification of the register under Section 50 of the Copyright Act, 1957, or adverse equitable consequences for the applicant. Accordingly, applicants should identify the protectable human contribution and avoid claiming rights in material that lacks sufficient human authorship.

Trade mark registration turns on proprietorship of a mark for the relevant goods or services, not on creative authorship. Design applications must accurately establish ownership and novelty. Patent applications must identify the true human inventor and establish the applicant’s entitlement. Naming an AI system as an inventor, or materially misstating inventorship, is likely to result in objection, opposition, revocation or an ownership dispute in the patent.

An AI system cannot be named as an inventor or co-inventor. The Patents Act, 1970 requires a “person” claiming to be the true and first inventor, an assignee or a legal representative, and the CRI Guidelines expressly reject autonomous AI inventorship while confirming that AI-assisted inventions are not excluded on that ground alone.

An AI-assisted invention can be patented where a human contributed to the inventive concept by defining non-obvious constraints, designing or interpreting experiments, selecting candidates for technical reasons, or integrating results into the claimed solution. Merely funding, owning or operating the AI system does not suffice for patentability.

Entitlement is determined by inventorship, assignment, employment and research-collaboration arrangements.

Where AI tools or automated experimentation are routinely used in the relevant technical field, they may be considered part of the notional skilled person’s ordinary technical resources. This may affect whether a claimed combination was obvious-to-try. The mere availability of AI will not, by itself, render every AI-generated result obvious. Enablement must continue to be assessed by reference to whether the disclosure enables the skilled person to perform the invention without undue experimentation.

An AI-generated publication, technical document, software code or synthetic dataset may constitute prior art if it was publicly available before the relevant priority date and contains a sufficiently clear and enabling disclosure. Human authorship is not necessarily required for publicly available material to form part of the state of the art. Conversely, confidential or inaccessible AI outputs, hallucinated or unreliable material, and disclosures that do not enable the invention will ordinarily not qualify as prior art.

Under the Designs Act, 2000, new or original features of shape, configuration, pattern or ornament applied to an article can be registered if such feature appeals solely to the eye and has not been previously published and is not dictated solely by function. Protection for Graphic User Interface (GUI), icons and virtual goods is case-specific and is dependent upon satisfying the applicable “article” and classification requirements.

An AI system cannot own or register a design. Under the Designs Act, 2000, ownership must vest in a human or legal entity through creation, employment, commissioning, assignment or contract. Under the Copyright Act, 1957, copyright protection ceases to be available for a design capable of registration under the Designs Act, 2000 if it has not been registered and is industrially reproduced on an article more than 50 times. Distinctive, source-identifying product appearance and packaging may be protected through trade mark registration under the Trade Marks Act, 1999.

AI-generated or AI-assisted names, logos, slogans, sounds, shapes and motion marks can be registered under the Trade Marks Act, 1999, provided that the mark:

• is capable of being represented graphically;
• possesses the requisite distinctiveness;
• identifies the commercial source of the relevant goods or services;
• does not conflict with a prior mark; and
• is not vague, deceptive or misleading.

Human authorship is not a prerequisite for trade mark registration. However, an AI system cannot, however, own a trade mark or apply for it unless a human is attributed with authorship under the Trade Marks Act, 1999. The proprietor of the trademark must be a legal person who has adopted, used or intends to use the mark. Ownership of an AI-assisted brand asset would typically be determined by the applicable contractual arrangements between the employer, agency, customer and AI-platform provider.

The Copyright Act, 1957 protects an author’s right to claim authorship and to restrain or claim restitutionary damages for any distortion, mutilation or modification prejudicial to the author’s honour or reputation. These protective rights would apply where AI modifies or manipulates a protected work or performance.

A wholly autonomous AI-generated output attracts no moral rights, because an AI system cannot be an author or a performer. Mere imitation of a creator’s general artistic style does not constitute an infringement of moral rights in the absence of the unauthorised use, reproduction or prejudicial treatment of an identifiable protected work or performance.

A false representation that a work was created, authorised or endorsed by a particular person can support claims for passing off, infringement of personality or publicity rights, and defamation. Additionally, the assignment of economic rights does not transfer or extinguish the author’s statutory moral rights.

India does not have a single, comprehensive statute governing publicity or personality rights. Protection for a person’s name, likeness and related personal attributes does not arise under a single legislation. Instead, the protection of a person’s likeness can be claimed under the fundamental rights to protection of life and personal liberty enshrined in the Constitution of India, the Trade Marks Act, 1999 and the Copyright Act, 1957, supplemented by the common law action for passing off.

The courts can restrain the unauthorised commercial use of an individual’s name, image, likeness, voice, persona or digital replica, particularly where such unauthorised commercial use implies endorsement, facilitates impersonation or creates a misleading association. Voice cloning and synthetic performances can also compromise a copyright or performers’ rights where they reproduce, incorporate or derive from protected recordings or performances.

Commercial courts and the high courts adjudicate AI-related intellectual property disputes. There is no AI-specific tribunal presently constituted in India. The appropriate forum is determined by:

• the nature of the right asserted;
• the value of the dispute;
• the cause of action; and
• the territorial nexus of the parties.

Civil infringement and passing-off suits are brought before the commercial courts and the commercial divisions of the high courts, while the intellectual property divisions of the high courts exercise jurisdiction over designated original proceedings, appeals and rectification or cancellation matters. The Supreme Court of India constitutes the final appellate forum.

Disputes regarding ownership, licensing and confidentiality can be referred to mediation or arbitration where the subject matter is arbitrable, although it is well settled, following the judgment in Vidya Drolia v Durga Trading Corporation, that an arbitral tribunal cannot conclusively determine rights in rem and can only ascertain rights in personam.

The Code of Civil Procedure, 1908, read with the Commercial Courts Act, 2015, prescribes the mechanisms for the discovery and preservation of evidence in such proceedings (including document disclosure, inspection, interrogatories, electronic discovery, preservation orders, the appointment of local commissioners and the drawing of adverse inferences in cases of spoliation or non-production).

To safeguard proprietary technical information and personal data, the courts routinely constitute confidentiality clubs, including under the Delhi High Court Intellectual Property Rights Division Rules, 2022, restrict access to counsel or nominated experts, order sealed filings and redactions, conduct proceedings in camera, appoint neutral technical experts, and permit inspection through sampling or within a secure environment.

The availability of interim or final injunctive relief in AI and intellectual property disputes depends on the nature of the right asserted, the impugned conduct, the stage of proceedings and the technical relationship between the allegedly infringing material and the relevant AI system.

The tripartite test governs interim relief:

• a prima facie case;
• the balance of convenience; and
• irreparable harm.

Courts typically consider proportionality, public interest, delay or acquiescence, the adequacy of damages, and the technical feasibility and enforceability of the relief sought.

Depending on the facts, the court can:

• direct the removal of identified works from datasets, repositories, or retrieval-augmented generation indexes;
• block specified outputs;
• require filters, guardrails, access controls, or human approval;
• suspend particular model versions; or
• disable autonomous publication and transaction functions.

The courts can also order disclosure or inspection of information concerning the sources of the allegedly infringing material, training or fine-tuning processes, output logs and safeguards implemented by the defendant.

Final relief typically includes permanent injunctions, the destruction or removal of infringing material and continuing compliance directions. The claimant would need to establish that the relevant conduct or artefact is actionable and that the relief sought is necessary, proportionate, technically feasible and enforceable.

Indian law does not presently recognise a settled equivalent of the “fruit of the poisonous tree” doctrine by which every downstream model, model weight or output is automatically tainted by allegedly unlawful training material. Accordingly, final relief is directed at a model’s internal state and is fact-specific, and would be contingent on proof of infringement, causation, the persistence or reproducibility of the protected material, continuing harm and the availability of less restrictive alternatives. Courts are therefore more likely to favour specific relief over wholesale deletion or reconstruction of an AI model.

A successful claimant can obtain permanent or interim injunctions, compensatory damages, delivery up or destruction of infringing material, costs and, where applicable, contractual remedies. India has no statutory damages regime. Damages require proof of loss and causation and may, where appropriate, be assessed by reference to the licence fee or reasonable royalty that would have been payable, and the courts award punitive damages in cases of flagrant infringement. An account of profits is limited to profits attributable to the infringement.

Intellectual property rights are territorial in nature. However, Indian courts can grant in personam relief against defendants within their jurisdiction, including orders requiring the defendant to take or refrain from taking action outside India. A foreign judgment obtained in a notified reciprocating territory may be executed directly in India under the Code of Civil Procedure, 1908. Where the judgment originates from a non-reciprocating territory, enforcement requires the judgment-creditor to file a fresh suit in India founded on the foreign judgment, subject to the sweeping defences to enforceability set out in the Code of Civil Procedure, 1908. However, foreign injunctions are not automatically recognised or enforceable in India.

An AI-content licence should clearly define the licensed content, authorised models and model versions, territory, term, permitted uses and whether the licence is exclusive. The licence terms should expressly address scraping, copying, pre-training, fine-tuning, retrieval-augmented generation, caching, synthetic-data generation and the reproduction or communication of the content through model outputs.

The agreement or licence terms should identify permitted affiliates, contractors, vendors and sublicensees, and regulate the use of agents, domains, APIs and autonomous publication or transaction capabilities.

The terms should also address ownership and permitted use of outputs, usage reporting, audit rights, confidentiality, information security, personal data, takedown procedures, deletion obligations and the technical limitations of machine unlearning.

The DPIIT Working Paper on Generative AI and Copyright (Part I), released in December 2025, proposed a mandatory, blanket licensing framework permitting the use of lawfully accessed content for AI training without individual negotiations, with remuneration collected on commercialisation through a centralised mechanism. Part II, which will address authorship and output-side questions, is awaited. The proposal is not presently law, and significant questions remain regarding royalty rates, the allocation and distribution of remuneration, opt-out rights, its application to foreign services, and its interaction with the subsequent decision in ANI Media Private Limited v OpenAI OpCo LLC.

The India AI Governance Guidelines, released by the MEITY in November 2025, favour an existing-law, risk-based framework centred on auditability, accountability and institutional co-ordination. The Guidelines for Examination of Computer Related Inventions, 2025 are shaping examination practice for AI-related patent applications. Separately, the draft Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Second Amendment Rules, 2026, released for consultation on 30 March 2026 and not yet notified, would make compliance with MEITY advisories and directions part of the statutory due diligence obligation on which safe harbour depends. Right-holders, AI developers and deployers should accordingly maintain comprehensive records regarding licensing arrangements, market impact, model development and agent controls.

India participates increasingly in international discussion of AI and intellectual property through WIPO, the WTO and the TRIPS framework, the G20, bilateral trade and digital-policy dialogues, and international standards organisations.

It has, however, taken on no binding international commitment specific to AI and intellectual property. India’s statutory framework regarding IP and AI differs from that of the European Union, which recognises a sui generis database right and provides express text-and-data-mining exceptions supported by a statutory, machine-readable opt-out mechanism.

It also differs from the United States, where the open-ended doctrine of fair use performs the work done in India by the enumerated fair-dealing exceptions in the Copyright Act, 1957. The Delhi High Court’s reading of the relevant section of the Copyright Act, 1957 – ie, Section 52(1)(a) in ANI Media narrows that gap at the interim stage, but the enumerated structure of the Indian exception remains unchanged.

The Copyright Act, 1957’s applicability, as a whole, remains judicially untested for generative AI output, and the royalty-based blanket licensing model proposed in the DPIIT Working Paper would, if enacted, be without close parallel in any major jurisdiction.