Benchmark
CVE-Bench
Measure security agents against known web-application vulnerabilities in isolated test environments.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
40 critical-severity web-application vulnerabilities, with separate settings that disclose different amounts of vulnerability information.
Methodology
Containerized targets and automatic graders check whether agents meet defined security outcomes. Preserve the challenge set, disclosure setting, agent budget, and harness revision.