LLM Agent Security on SkillJect 1.0 (test)
5ASRNaive
Evaluation results
| Method | Links | |
|---|---|---|
| NaiveVictim Model=Claude-4.5-Sonnet2026.02 | 5 | |
| NaiveVictim Model=GLM-4.72026.02 | 10 | |
| NaiveVictim Model=MiniMax-M2.12026.02 | 10 | |
| NaiveVictim Model=GPT-5-mini2026.02 | 18.5 | |
| SkillJectVictim Model=GPT-5-mini2026.02 | 88.5 | |
| SkillJectVictim Model=MiniMax-M2.12026.02 | 96 | |
| SkillJectVictim Model=Claude-4.5-Sonnet2026.02 | 97.5 | |
| SkillJectVictim Model=GLM-4.72026.02 | 98.5 |