Benchmark
AILuminate Jailbreak
Test whether AI safeguards withstand adversarially rewritten requests.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The October 29, 2026 Candela announcement identifies Jailbreak 1.1 across eleven hazard categories. The broader benchmark family covers text and image-plus-text attacks.
Methodology
Compare behavior under attacks with ordinary safety testing. The family describes a Resilience Gap; the Candela launch describes classifier grading, versioned results and separate practice and official question sets. The overview links older white papers, so modality and edition must remain explicit.