phenomenon · first seen 8 Oct 2026
AI safety evaluation of model jailbreak robustness
Also: HarmBench, StrongREJECT
Measuring how inference optimizations alter a model's refusal and harmful-output behavior compared with its native baseline.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Searching for "Harmful Refusal": A Psychometric Audit of an AI Safety BenchmarkPapers · 8 Oct 2026 · Christopher M. Stewart, Preston Botter, Natalie Sarabosing, Muye Zhang, Rachel P…
- To Cache or Not to Cache: Understanding and Mitigating Safety Incompatibility in Cross-Model KV Cache ReusePapers · 11 Oct 2026 · Zhiwei Lin, Jiahao Liu, Yan Liu, Chengpeng Wang
Latest sources
- To Cache or Not to Cache: Understanding and Mitigating Safety Incompatibility in Cross-Model KV Cache ReusePapers · 11 Oct 2026 · Zhiwei Lin, Jiahao Liu, Yan Liu, Chengpeng Wang
- Searching for "Harmful Refusal": A Psychometric Audit of an AI Safety BenchmarkPapers · 8 Oct 2026 · Christopher M. Stewart, Preston Botter, Natalie Sarabosing, Muye Zhang, Rachel P…