Tag
This paper proposes a methodology for deriving harmonized AI safety thresholds across frontier AI companies to address inconsistencies in existing thresholds, covering misuse risks and automated AI R&D, and highlighting empirical gaps.