Collegium Helveticum
66cf35f9dbf5de5f5acd6d4d_662b7e4f60bd0d5befaaf1c8_back-box-decor
Wendy K. Tam aims at opening up the "black box" of an AI model to see how it internally represents concepts. AI generated by Eduard Goricev.
Fellow Project 2026–2027

From Platforms to Models
Internet Governance in the Age of Generative Intermediaries

As large language models increasingly replace search engines as our primary gateway to knowledge, this project asks: can legal systems built to regulate human-authored content still govern systems that generate it?

Wendy K. Tam's current research examines the legal foundations of internet governance, emphasizing how distinct legal epistemologies in the United States and European Union shape intermediary liability, speech, and data protection, and how law and computation jointly produce social outcomes. The project extends this inquiry to the generative realm, as large language models displace search engines and platforms as the primary interface between citizens and knowledge. This shift unsettles regulatory frameworks designed for intermediaries that retrieve human-authored content rather than systems that generate it.

Wendy uses mechanistic interpretability of open-weight models to demonstrate that political framing is encoded as a locatable and steerable geometric direction within the model, and that alignment training masks rather than removes this structure. These tools render a generative intermediary's internal choices legible, precisely where legal measurement premised on identifiable sources breaks down. The aim is a globally grounded account of how legal systems and model architectures will jointly govern the emerging information environment.