AI Policy Researcher Calls for Public Guidance on Defining 'Frontier' Models

evijit · x · 2026-08-13

Responding to reports that the US government plans to expand AI oversight to open-source frontier models, the author argues that from both policy and evals research perspectives, clear and public guidance is needed on what constitutes a 'frontier' model. While acknowledging Goodhart's law, the author emphasizes that public policy should be made with public evals wherever possible.

Original post →

More from Safety

Safety channel →