Trade Secret Protection of AI Models: Balancing Innovation Incentives and Algorithmic Transparency

Zeyu Li

doi:10.54691/sbrdam63

Authors

Zeyu Li Edinburgh University, Edinburgh, United Kingdom

DOI:

https://doi.org/10.54691/sbrdam63

Keywords:

Trade secret, AI model, algorithmic transparency, innovation incentive, confidential disclosure.

Abstract

Developers of cutting-edge AI models are increasingly using trade secrecy to protect these models rather than patents; disclosure would reveal the configuration that gives the model its value, and thus secrecy also obstructs the algorithmic transparency that regulators, courts, and affected individuals now demand. This paper regards this tension as a boundary issue of a single intellectual property right rather than a conflict of incompatible goods. The European Union, the United States and China cut transparency obligations into trade secret protection; in each, disclosure stops short of the public rather than reaching it. Building on this convergence, this paper distinguishes between disclosure that only weakens a secret at the margin and public disclosure that destroys it completely, and holds that the current settlement, due to procedural asymmetry, draws the line too broadly. Calibrated confidential disclosure would relocate it defensibly.

Downloads

Download data is not yet available.

References

[1] Hrdy, C. A., & Lemley, M. A. (2021). Abandoning trade secrets. Stanford Law Review, 73, 1.

[2] Campanelli, G. L. (2024). Can ChatGPT keep a secret? An evaluation of the applicability and suitability of trade secrecy protection for AI-generated inventions. Duke Law & Technology Review, 24, 1.

[3] Sandeen, S. K., & Aplin, T. (2022). Trade secrecy, factual secrecy and the hype surrounding AI. In Research handbook on intellectual property and artificial intelligence (pp. 443–460). Edward Elgar Publishing.

[4] Levine, D. S. (2023). Generative artificial intelligence and trade secrecy. Journal of Free Speech Law, 3, 559.

[5] Mylly, U. M. (2023). Transparent AI? Navigating between rules on trade secrets and access to information. IIC-International Review of Intellectual Property and Competition Law, 54(7), 1013–1043.

[6] Popowicz-Pazdej, A. (2023). The proportionality between trade secret and privacy protection: How to strike the right balance when designing generative AI tools. Journal of Data Protection & Privacy, 6(2), 153–166.

[7] Matulionyte, R., & Aranovich, T. (2022). Trade secrets versus the AI explainability principle. In Research handbook on intellectual property and artificial intelligence (pp. 405–422). Edward Elgar Publishing.

[8] Xu, J. (2024). Opening the 'black box' of algorithms: Regulation of algorithms in China. Communication Research and Practice, 10(3), 288–296.

[9] Lemley, M. A. (2008). The surprising virtues of treating trade secrets as IP rights. Stanford Law Review, 61, 311.

[10] Bucknall, B., Trager, R. F., & Osborne, M. A. (2025). Position: Ensuring mutual privacy is necessary for effective external evaluation of proprietary AI systems. arXiv:2503.01470.

[11] Wexler, R. (2018). Life, liberty, and trade secrets: Intellectual property in the criminal justice system. Stanford Law Review, 70, 1343.

[12] Kewanee Oil Co. v. Bicron Corp., 416 U.S. 470 (1974).

[13] State v. Loomis, 881 N.W.2d 749 (Wis. 2016).

[14] Trinidad v. OpenAI, Inc., No. 25-cv-06328-JST (N.D. Cal. Jan. 5, 2026).

[15] Food Marketing Institute v. Argus Leader Media, 139 S. Ct. 2356 (2019).