Copyright, Trade Secret or Agreement? Intellectual Property Framework for Proprietary Prompts

Introduction : The rise of the commercial side of generative artificial intelligence brought about a unique intellectual asset, prompt. Prompt became not a spontaneous message to be entered into a chatbot, but an elaborate set of instructions tested in a loop until it yielded the required result from the large language model. Enterprises develop prompt libraries which are considered their proprietary assets, comparable to source code in terms of internal business significance, while websites like PromptBase have already formed the market of tradable prompts, where prompt engineers can earn money on selling their individual prompts.  

Such a commercial reality clashes with the current intellectual property regime. First, the copyright regime provides for the protection of literary expression and not functional instruction; second, trade secrets protect information that is kept in confidence, but the worth of a prompt lies in its disclosure to a third-party AI, which can then be deconstructed to find out the functional instruction contained in it; and finally, contract law is limited to the contracting parties. The following discussion critically analyses the relevance of copyright, trade secrets, and contracts in relation to prompt libraries, function of prompts, and reverse engineering.  

A. Copyright Law and the Threshold of Originality

A copyright exists in an original literary work as per Section 13(1)(a) along with Section 2(o) of the Copyright Act, 1957, wherein "literary work" has been defined to mean compilations and computer programmes. The requirement here is the "modicum of creativity" standard as laid down by the Supreme Court in Eastern Book Company v. D.B. Modak, which entails something beyond mechanical effort, without requiring a literary quality. No person shall be entitled to a copyright except under the provisions of the Act as provided for in Section 16. Thus, there can be no common law claim to copyright/ based on the text of the prompt.  

The corresponding clause in the American law, Section 102(b) of the Copyright Act, 1976 (17 U.S.C. § 102(b)) explicitly states that copyright protection is not available for ideas, procedures, processes, systems, and methods of operations. Since a prompt is essentially a procedure, this exclusion is directly relevant in cases where originality in the prompt is called into question.  

B. The Stance of the United States Copyright Office Regarding Prompts

Prompting is one aspect of AI art addressed in the U.S. Copyright Office’s Report entitled Copyright and Artificial Intelligence, Part 2 : Copyrightability, published on 29 January 2025. It was argued by the Office that even when text prompts are used in great detail or repeatedly refined during the process, it does not give rise to copyright in the output because they serve as instructions for implementing ideas which are not copyrightable. The fact that AI models behave unpredictably in generating output prevents the establishment of a causal connection between the user’s input and the expressive aspects of the output produced. However, it is still possible that a prompt, as text on its own, is capable of being protected by copyright due to literary aspects in its wording. 

C. Trade Secrets and Confidentiality Laws 

In the United States, protection of trade secrets is covered by the Uniform Trade Secrets Act, adopted by many states and the Defend Trade Secrets Act of 2016. In each case, trade secrets are defined as information which, because of its confidentiality, provides the owner with independent economic value, and their misappropriation occurs when such information has been acquired through "improper means" or has been disclosed in breach of a duty of confidentiality, but neither act makes independent acquisition through reverse engineering unlawful.  

In India, there is no specific law covering trade secrets. Protection of trade secrets is based on the common law principle of the breach of confidence doctrine and the law of contracts, but Section 27 of the Indian Contract Act of 1872 precludes any agreement that restrains a person from exercising his lawful profession, trade or business, with the sole exception of sale of goodwill. Despite this restriction, Indian courts have enforced confidentiality obligation independent of post-employment non-competition obligation, and recognised that a person can be restrained from using compilations of confidential information even though a noncompetition agreement is unenforceable.  

The Originality and Merger Issue in Individual Prompts

The initial problem in the copyright protection of an individual prompt is associated with the distinction between idea and expression made in Baker v. Selden - copyright covers only a particular expression of an idea and not the idea, system or method. If an idea is capable of only a few expressions, the principle of merger applies and makes both idea and expression unprotected because protection of expression would mean a grant of monopoly for the idea. Some functional prompts are likely to be subject to analysis through the prism of merger: there is only a limited set of ways how to express the functional requirement and protection of any expression would preclude competition in instructions for the performance of the function.

The decision of the Court of Justice of the European Union in SAS Institute Inc. v. World Programming Ltd provides a good comparison. According to the decision, the source code of a computer programme can be protected expression but the functionality of a programme and its data-file formats are not protectable expressions.  

Compilation Copyright in Curated Prompt Libraries

Although individual prompts themselves are constrained functionally, a library of prompts, as an entire collection, may fall within the category of compilation copyright where the selection, arrangement, and categorization of prompts involves some editorial judgement. The principle stated in Feist Publications, Inc. v. Rural Telephone Service Co. comes into play at this point: even though facts, and functional aspects, are not eligible for copyright, a compilation can be protected under copyright due to the fact that it is original and involves the creative selection and arrangement of facts or ideas; the arrangement must not be so trivial and automatic that there is no room left for any creative freedom.  

Trade Secrets and the Threat of Reverse Engineering via Output Analysis

In principle, trade secret law appears to be a good match for prompts since it does not demand originality or fixation but only the economic value stemming from secrecy and adequate precautions. The growing practice of treating structured libraries of prompts as trade secrets is explained by that. The legal framework, however, faces a challenge which is peculiar to prompts - the possibility for an attacker, who has no access to the text of the prompt, to deduce its contents via output analysis, which has been shown to be possible in the literature on prompt stealing attacks, where researchers showed that carefully crafted “shadow” queries could reconstruct the semantic content of an unknown prompt with considerable precision. This generates a conflict in doctrine.

The doctrine of trade secrets permits the lawful activity of reverse engineering, which entails a legal act of redeveloping any protected information based on the analysis of a legally acquired product, as a complete defense against any misappropriation based on the understanding that once the product enters the market without proper protection, the internal mechanisms are subject to analysis. If the process of prompt input/output analysis is considered a lawful method of analysis of a publicly available AI service, then the underlying prompt will cease to have trade secret protection once its output becomes publicly known because the information can be "readily ascertainable by proper means." On the other hand, when the access control measures are bypassed and injection of prompts occurs to induce the disclosure of the instructions, then such behavior has been characterized as an "improper means."

Contractual Obligation as the Residual Mode of Protection

Considering that copyright does not provide sufficient protection to functional prompt text and the natural erosion of secrecy resulting from output-oriented deployment, it remains the most reliable form of protection to rely on the confidentiality obligation. Confidentiality obligations independent of copyright have been enforced by Indian courts in cases where proprietary compilation of the material was at issue: in Mr. Diljeet Titus, Advocate v. Alfred A. Adebare, the Delhi High Court issued an order restraining the former associate from using client databases and proprietary compilation performed while working with the client. In Burlington Home Shopping Pvt. Ltd. v. Rajnish Chibber, an injunction was imposed to restrain the use of compiled customer databases without copyright infringement. 

The limitation on confidentiality clauses comes from Section 27 of the Indian Contract Act, 1872, which provides that a confidentiality clause that restrains the use and/or disclosure of proprietary materials specified by the contract is valid, while a confidentiality clause that would limit a former employee or licensee from engaging in the profession of prompt engineering shall be treated as an unenforceable restraint of trade.

Case Laws Applicable

Baker v. Selden, 101 U.S. 99 (1879) (United States): Introduced the doctrine of ideaexpression dichotomy whereby describing a system cannot vest copyright in the system, and is hence relevant to prompt instructions which are functional.   

Eastern Book Company & Ors. v. D.B. Modak & Anr., (2008) 1 SCC 1 (India): Applied the "modicum of creativity" test for originality, requiring skill and judgment over and above mechanical work.

Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991) (United States): It was decided that copyright in compilations exists only in its selection and arrangement, hence relevant to compilation-based copyright protection for libraries of prompts. 

SAS Institute Inc. v. World Programming Ltd, Case C-406/10, [2012] ECR I-000 (CJEU): It was decided that functionality and data format were not expressions, and thus an analogy can be drawn between non-protectable prompt logic and protected prompt phrasing.  

Mr. Diljeet Titus, Advocate v. Alfred A. Adebare & Ors., 130 (2006) DLT 330 (Delhi High Court, India): Prohibition on misappropriation of confidential databases and compilations of work irrespective of copyright.   

Burlington Home Shopping Pvt. Ltd. v. Rajnish Chibber, 61 (1995) DLT 6 (Delhi High Court, India): It awarded injunctive relief restraining the use of a compiled customer database based on confidentiality.

Practical Implications

Since there does not exist any intellectual property right by itself that would sufficiently protect the prompt library, a multi-layered approach should be adopted by businesses and individuals treating their prompts as confidential material.  

Firstly, it is necessary to prepare non-disclosure agreements with utmost clarity. The term "confidential information" within such agreements relating to the access of the library of prompts must clearly include the prompt, chain of thought structure, and parameters set for the prompts, along with appropriate use restrictions beyond the term of the agreement, so that the agreement continues to be enforceable despite Section 27 of the Indian Contract Act.  

Secondly, the licensing agreement must be clear about the scope of the grant: whether the license is exclusive or non-exclusive, field of use and source of models limitation, whether the licensee can sub-license the material or not, and whether the licensee gets any rights over the output generated using the licensed prompts. The ownership of outputs generated using the prompt becomes particularly important to clarify, since the issue of ownership of AI-generated outputs is itself a gray area; hence, it is best if the licensing agreement clarifies the point. Third, access control and technical measures need to mitigate the risk of output-based reverse engineering, such as query rate limits, forbidding bulk or automated queries via the terms of service, and even watermarking to identify unauthorized use of a prompt’s outputs, which has been studied recently by means of watermark injection and verification. The terms of service must explicitly define prompt injection in order to extract instructions as an improper form of access, which would strengthen any claim of trade secrets.  

Fourth, discipline needs to be upheld in the handling of evidence: the history of versions, drafts with timestamps, and the documentation of access restrictions are relevant in establishing both originality and the "reasonable measures" requirement for trade secrets. 

Lastly, where copyright of compilations is applicable, formal registration of the compilation needs to be sought as well, despite the fact that registration is helpful but will not cover the functional aspects of the individual prompts within the compilation.

Conclusion

The current state of prompt libraries in relation to intellectual property doctrine is a difficult one. The Copyright Act, limited as it is by the idea-expression distinction and the merger doctrine, provides only limited protection to functional individual prompts, although some protection might conceivably be available for the creative selection and organization of a library of prompts. Trade secret law, on its face a doctrine well suited for prompt protection, suffers from the same structural weakness of output-based reverse engineering inherent in deployment against a model which reveals information from which a prompt may be ascertained.

The remaining confidentiality obligation thus stands out as the most solid, though not necessarily perfect, means of protection, always subject to antitrust restraints preventing a confidentiality agreement from being a subterfuge for a non-compete covenant.  

In the absence of legislation or judicial guidance on how prompt protection should be tailored, the appropriate response appears to be a combination of efforts to obtain compilation copyright if curatorial skill is shown, retain trade secret status via controlled access if possible, and rely mostly on carefully drafted confidentiality and licensing agreements between the stakeholders involved with the prompt.  

Author: Ajit Tada in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at  Khurana & Khurana, Advocates and IP Attorney

Endnotes

1. Copyright Act, 1957, s. 13(1)(a) read with s. 2(o) (India).

2. Copyright Act, 1957, s. 16 (India).

3. Copyright Act, 1976, 17 U.S.C. § 102(b) (United States).

4. United States Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (29 January 2025).

5. Uniform Trade Secrets Act (as adopted by U.S. states); Defend Trade Secrets Act of 2016, 18 U.S.C. § 1836 et seq. (United States).

6. Indian Contract Act, 1872, s. 27 (India).

7. Baker v. Selden, 101 U.S. 99 (1879) (United States).

8. Eastern Book Company & Ors. v. D.B. Modak & Anr., (2008) 1 SCC 1 (India).

9. Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991) (United States).

10. SAS Institute Inc. v. World Programming Ltd, Case C-406/10, [2012] ECR I-000 (Court of Justice of the European Union).

11. Mr. Diljeet Titus, Advocate v. Alfred A. Adebare & Ors., 130 (2006) DLT 330 (Delhi High Court, India). 

12. Burlington Home Shopping Pvt. Ltd. v. Rajnish Chibber, 61 (1995) DLT 6 (Delhi High Court, India).

13. Y. Shen et al., 'Prompt Stealing Attacks Against Text-to-Image Generation Models,' arXiv:2302.09923 (2023).