Malicious Behavior Induction on Skillsmp randomly sampled 300 skills
52Credential Harvesting ASRDyMalSkill
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DyMalSkillAgentic Framework=OpenHands, Backbone LLM=GPT-4o-mini2026.06 | 52 | 32 | 32 | 52 | 42 | 44 | 30 | 40 | 30 | 70 | 10 | 68 | 41.8 | |
| DyMalSkillAgentic Framework=OpenHands, Backbone LLM=GPT-52026.06 | 36 | 38 | 20 | 40 | 30 | 40 | 32 | 36 | 42 | 64 | 66 | 80 | 43.7 | |
| DyMalSkillAgentic Framework=OpenHands, Backbone LLM=Qwen3.6-35B2026.06 | 25 | 26.3 | 44 | 37 | 38.3 | 42.3 | 47.3 | 29 | 28 | 65 | 44.3 | 72.7 | 41.6 | |
| DyMalSkillAgentic Framework=OpenHands, Backbone LLM=Qwen3-8B2026.06 | 19.3 | 14.7 | 12 | 11.7 | 11.7 | 13.7 | 21.3 | 13.3 | 6 | 40.7 | 20.7 | 39 | 18.7 | |
| DyMalSkillAgentic Framework=Claude Code, Backbone LLM=Sonnet 4.62026.06 | 12 | 4 | 0 | 36 | 0 | 32 | 46 | 2 | 0 | 4 | 10 | 82 | 19 |