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

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.