Tool
MLCommons Jailbreak Benchmark v1.0
First reported · Discovered arxiv.org
Page published
Earliest dated coverage: 29 Oct 2026 · First observed: 4 Nov 2026 · Latest dated coverage: 29 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
AILuminate gives defenders repeatable tests for comparing AI systems’ safety behavior and resistance to jailbreaks before deployment.
MLCommons released AILuminate Safety 1.2.1 and AILuminate Jailbreak 1.1 for on-demand testing through Candela, alongside public benchmark results for 27 AI systems. The benchmarks assess harmful-output handling and resistance to adversarial prompts, with the supporting jailbreak methodology combining baseline and adversarial evaluations, human annotation, and automated evaluator calibration.