S3 composes stage-specific safety skills through a guard agent, achieving near-zero attack success on six risk types in its own benchmark while preserving benign task completion.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
$S^3$: Improving Agent Safety through Multi-Stage Defense
S3 composes stage-specific safety skills through a guard agent, achieving near-zero attack success on six risk types in its own benchmark while preserving benign task completion.