Using item response theory on 192 models and eight safety benchmarks, this paper finds three latent safety factors, cuts evaluation cost by 97-99% with adaptive item selection, and detects naive sandbagging and API model changes.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Item Response Theory for AI Safety
Using item response theory on 192 models and eight safety benchmarks, this paper finds three latent safety factors, cuts evaluation cost by 97-99% with adaptive item selection, and detects naive sandbagging and API model changes.