AI Safety Frameworks Should Include Procedures for Model Access Decisions

Edward Kembery, Tom Reed

The downstream use cases, benefits, and risks of AI models depend significantly on what sort of access is provided to the model, and who it is provided to. Though existing safety f…