korshunov.ai
English
Benchmark · safety

JailbreakBench

2 results 2 models
0 24.5 49 73.5 98 2026-07-04 STEER (applied to six open-source 8B-parameter models) · 93 · 2026-07-04 GPT-4o-mini · 35.5 · 2026-07-04
STEER (applied to six open-source 8B-parameter models) GPT-4o-mini
Timeline
  1. 2026-07-04 STEER (applied to six open-source 8B-parameter models) 93.0% STEER attack exposes LLM safety gaps in low-resource languages
  2. 2026-07-04 GPT-4o-mini 35.5% STEER attack exposes LLM safety gaps in low-resource languages