Sefz discovers specification violations in 29.9% of 402 real-world agent skills by translating guardrails into reachability goals and guiding LLM mutations with a multi-armed bandit.
Here comes the ai worm: Preventing the propagation of adversarial self-replicating prompts within genai ecosystems
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
No Attack Required: Semantic Fuzzing for Specification Violations in Agent Skills
Sefz discovers specification violations in 29.9% of 402 real-world agent skills by translating guardrails into reachability goals and guiding LLM mutations with a multi-armed bandit.