A preprint proposes six threat-actor attributes to make pre-release misuse evaluations of open-weight frontier AI more consistent.