A compilable, multi-task, mutation-aware Rust vulnerability benchmark finds frontier agents at 56–65% detection but ~20% line F1, with adversarial cues dropping line F1 ~27%.
Proceedings of the 17th European Workshop on Systems Security , pages =
1 Pith paper cite this work, alongside 2 external citations. Polarity classification is still indexing.
1
Pith paper citing it
2
external citations · external index
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RustMizan: A Compilable, Contamination-Aware Benchmarking Framework for Rust Vulnerabilities
A compilable, multi-task, mutation-aware Rust vulnerability benchmark finds frontier agents at 56–65% detection but ~20% line F1, with adversarial cues dropping line F1 ~27%.