Preprint Proposes Clearer AI Threat Profiles Before Release
A preprint proposes six threat-actor attributes to make pre-release misuse evaluations of open-weight frontier AI more consistent.
Developing Light ยท https://developinglight.com/editorial/developing-light