An online KS-statistic monitor detects shifts in deployed safety classifiers with 86.6% valid detection rate, exposes conformal prediction collapse in high-dimensional embeddings, and derives a confidence-gated security boundary against adaptive attackers.
I can't believe it's not robust: Catastrophic collapse of safety classifiers under embedding drift
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Online Shift Detection and Conformal Adaptation for Deployed Safety Classifiers
An online KS-statistic monitor detects shifts in deployed safety classifiers with 86.6% valid detection rate, exposes conformal prediction collapse in high-dimensional embeddings, and derives a confidence-gated security boundary against adaptive attackers.