Intellectual property in synthetic datasets

Introduction

Synthetic datasets are artificially generated collections of data created through computer algorithms, artificial intelligence models, statistical simulations and computational techniques. Unlike traditional datasets, which contain information collected directly from real individuals, organisations or physical events, synthetic datasets reproduce selected statistical patterns and characteristics without necessarily containing actual personal records.

These datasets are increasingly used in biotechnology, healthcare, pharmaceutical research, artificial intelligence, financial modelling and scientific experimentation. For example, synthetic medical datasets can simulate patient records, disease patterns, genetic variations and clinical trial outcomes for research and algorithm development.

The creation and commercial use of synthetic datasets raise important intellectual property questions concerning copyright, database protection, ownership, originality, trade secrets and licensing. Legal difficulties arise when synthetic datasets are generated using copyrighted source material, confidential databases or proprietary AI models.

Therefore, intellectual property law must balance the protection of creators' investments with the need to encourage scientific research, technological innovation and responsible data sharing.

Legal and regulatory framework

1. Copyright Act, 1957

The Copyright Act, 1957 protects original literary, artistic, musical and other qualifying works.

Synthetic datasets may receive copyright protection where their selection, arrangement or presentation involves sufficient originality.

Section 2(o) includes compilations and computer databases within the definition of literary work.

Section 14 establishes the exclusive rights of copyright owners, including rights relating to reproduction and adaptation.

However, copyright does not automatically protect individual facts, raw data or every automatically generated dataset.

2. Patents Act, 1970

The Patents Act, 1970 protects inventions satisfying the requirements of novelty, inventive step and industrial applicability.

A synthetic dataset itself will not ordinarily qualify for patent protection merely because it contains artificially generated information.

However, a novel technical process for generating or using synthetic datasets may potentially qualify if it satisfies the statutory requirements and is not excluded under Section 3.

3. Information Technology Act, 2000

The Information Technology Act, 2000 provides legal protection against specified forms of unauthorised access, copying, damage and misuse of computer systems.

Section 43 may become relevant where synthetic datasets or their underlying databases are accessed, downloaded or copied without authorisation in circumstances covered by the provision.

Section 66 may apply where the relevant acts are committed dishonestly or fraudulently.

4. Digital Personal Data Protection Act, 2023

The Digital Personal Data Protection Act, 2023 regulates the processing of digital personal data in India.

Synthetic datasets may reduce privacy risks when they do not identify individuals. However, a dataset that can be linked back to identifiable individuals may still raise data protection concerns.

Organisations must assess whether the information qualifies as personal data under the Act rather than assuming that artificially generated data is automatically exempt.

5. Contract and trade secret protection

Synthetic dataset ownership may also depend on contractual arrangements between researchers, software developers, universities and commercial organisations.

Confidential generation methods, proprietary algorithms and unpublished datasets may be protected through non-disclosure agreements, licensing arrangements and contractual confidentiality obligations.

Major intellectual property issues in synthetic datasets

1. Copyright protection and originality

Synthetic datasets may contain millions of artificially generated records.

Copyright protection depends on whether the dataset constitutes an original compilation or another protected work.

For example, a biotechnology company may develop an original synthetic genomic dataset through substantial human selection, arrangement and organisation.

However, individual statistical values and factual information are not automatically protected by copyright.

2. Ownership of synthetic datasets

Ownership becomes complicated when datasets are generated using multiple technologies and contributors.

For example, a pharmaceutical company may use a licensed AI model, university research data and proprietary computational software to generate synthetic clinical trial records.

Questions may arise concerning ownership of the final dataset, underlying algorithms and associated intellectual property.

Ownership should therefore be addressed through clear contractual arrangements.

3. Copyright infringement in training datasets

AI systems used to generate synthetic datasets may be trained using copyrighted books, research papers, photographs, scientific databases or other protected material.

If copyrighted material is reproduced or used without proper authorisation, copyright infringement questions may arise depending on the nature of the use and applicable legal exceptions.

The fact that the final dataset is synthetic does not automatically eliminate possible infringement involving the source material.

4. Trade secret protection

Synthetic datasets may contain valuable commercial information, particularly when developed through confidential research and specialised computational methods.

Businesses may protect proprietary datasets through:

Confidentiality agreements.

Restricted access controls.

Licensing arrangements.

Employee confidentiality obligations.

Cybersecurity safeguards.

5. Database ownership and licensing

Organisations may invest considerable resources in generating and organising synthetic datasets.

Disputes may arise concerning commercial reuse, redistribution, modification and sublicensing.

Licensing agreements should specify permitted uses, attribution requirements, redistribution rights and restrictions on competing commercial exploitation.

6. Privacy and re-identification

Synthetic datasets are sometimes designed to avoid exposing actual personal information.

However, certain generation methods may reproduce identifiable patterns or allow individuals to be inferred through comparison with external datasets.

Therefore, privacy assessment and technical validation remain important even when data is described as synthetic.

Important judicial decisions

Indian courts have not yet developed a specialised body of case law exclusively concerning intellectual property rights in synthetic datasets. However, the following decisions establish important principles concerning database copyright, originality, confidential information and intellectual property protection.

1. Eastern Book Company v. D.B. Modak (2008)

The Supreme Court examined the originality requirement under Indian copyright law. It adopted a standard requiring sufficient skill and judgment rather than mere labour or mechanical effort.

Relevance: The judgment is important when determining whether a synthetic dataset's selection, arrangement or organisation involves sufficient originality to receive copyright protection. Merely generating large quantities of data automatically does not necessarily establish copyright ownership.

2. Burlington Home Shopping Pvt. Ltd. v. Rajnish Chibber (1995)

The Delhi High Court considered copyright protection relating to a commercially developed database and its unauthorised use.

Relevance: Synthetic datasets may involve substantial investment in collecting, selecting, organising and structuring information. The judgment provides useful principles for protecting qualifying database compilations against unauthorised copying.

3. American Express Bank Ltd. v. Priya Puri (2006)

The Delhi High Court examined the distinction between confidential business information and general knowledge or experience acquired by employees.

Relevance: Companies generating proprietary synthetic datasets may face risks when employees or former employees disclose confidential information. The judgment is relevant to determining whether particular datasets and associated business information qualify for confidentiality protection.

4. John Richard Brady v. Chemical Process Equipments P. Ltd. (1987)

The Delhi High Court examined the protection of confidential technical information against unauthorised use.

Relevance: Synthetic dataset generation may involve proprietary algorithms, technical procedures and confidential scientific information. The judgment provides principles relevant to protecting such confidential commercial and technical assets.

5. R.G. Anand v. Deluxe Films (1978)

The Supreme Court examined the distinction between protected expression and unprotected ideas.

Relevance: Synthetic datasets may reproduce general concepts, statistical relationships or scientific facts found in existing databases. The judgment helps explain why copyright protection generally concerns original expression rather than ownership of underlying ideas, facts or information.

Major legal and regulatory challenges

Intellectual property protection for synthetic datasets involves several challenges:

Uncertain ownership: Multiple researchers, organisations and AI developers may contribute to dataset generation.

Originality requirements: Automatically generated information may not satisfy copyright originality requirements.

Source-data infringement: Synthetic datasets may be generated using copyrighted or confidential source material.

Database protection: Indian copyright law does not establish a separate sui generis database right equivalent to the European Union's database right.

Trade secret disputes: Unauthorised disclosure of confidential generation methods may cause commercial losses.

Privacy concerns: Synthetic information may retain identifiable characteristics of real individuals.

Licensing disputes: Unclear contractual terms may create disagreements concerning commercial reuse and redistribution.

International jurisdiction: Synthetic datasets may be generated, stored and commercialised across multiple countries with different IP laws.

Recommendations for strengthening intellectual property protection

India should develop clearer guidelines concerning ownership, licensing and copyright protection for synthetic datasets.

Important measures should include:

Establishing transparent dataset ownership agreements.

Maintaining documentation of source data and generation methods.

Identifying human contributions to dataset creation.

Conducting copyright and confidentiality assessments.

Implementing technical safeguards against re-identification.

Establishing clear commercial licensing conditions.

Protecting proprietary algorithms and generation processes.

Encouraging responsible synthetic data sharing for scientific research.

Developing specialised guidance for AI-generated databases.

Research institutions and biotechnology companies should also maintain records identifying the source material, software licences, human contributors and contractual rights associated with every major synthetic dataset.

Conclusion

Intellectual property in synthetic datasets is an emerging legal issue at the intersection of artificial intelligence, copyright, database protection, data privacy and biotechnology. Synthetic datasets offer significant opportunities for scientific research, medical innovation and commercial development while reducing dependence on certain categories of real-world personal information. However, questions concerning originality, ownership, source-data infringement and trade secret protection remain important legal challenges. India's existing Copyright Act, Patents Act, Information Technology Act and data protection framework provide a foundation for addressing these issues. Clear ownership agreements, responsible data generation practices and stronger regulatory guidance are necessary to protect innovation while encouraging lawful access to synthetic datasets.