List your company

Prefill Attack on HEx-PHI Harmful

95.76Attack Success RateSafeLM
2.78426.92251.0675.198Jun 17, 2026

Evaluation results

MethodLinks
SafeLMJudge=Heuristic2026.0695.76
BaselineJudge=Heuristic2026.0692.42
BaselineJudge=LLM2026.0665.15
SafeLMJudge=LLM2026.0660.91
SRPJudge=Heuristic2026.068.48
SRPJudge=LLM2026.066.36