Foundation Model Transparency

The effective governance of foundation models is a clear challenge for global AI policy. Corporate transparency is a key lodestone: having sufficient information provided by the major AI companies to other stakeholders is necessary to make sensible decisions and forge sound policy. I will cover two works on foundation model transparency: (i) the annual Foundation Model Transparency Index, which scores major AI companies for their public-facing transparency and (ii) the current EU AI Act Code of Practice process, focusing specifically on the section on transparency. Overall, my talk will cover four themes: (i) how should we conceptualize transparency, (ii) what mechanisms influence transparency in the ecosystem, (iii) what is the current landscape of transparency, and (iv) what remains as a problem and how might we address it.