Cascading Secrecy: Tensions between IP and Transparency Objectives

Description

This paper explores how secrecy compounds across the AI supply chain and its implications for the public interest and innovation.

It was presented at the GigaNet Symposium of the 2025 UN Internet Governance Forum.

Download

• Preprint (SSRN)

Abstract

Secrecy is a feature, not a bug, for generative AI models. We explore this as a regime of successive interrelations that involves i. economic incentives towards secrecy based on high resource costs; ii. legal protections through intellectual property and the doctrine of trade secrets; and iii. policy concerns around public interest and tensions with technological innovation. A clear example of this regime at work can be seen in recent EU AI regulations, which seek to both promote transparency ideals whilst protecting trade secrecy. A proposed solution is sustained public investment in the governance and transparency of AI models, though such investment would need to be made across multiple layers of the generative AI model stack.