Adversarial Attack on LIAR ICL NEW
30.35Attack Success RateAgentic Adversarial Rewriting (Full History)
Evaluation Results
| Method | Links | |
|---|---|---|
| Agentic Adversarial Rewriting (Full History)Model Access=Direct Target Access, Optimization=Prompt Optimization, History=Full History2026.04 | 30.35 | |
| Agentic Adversarial Rewriting (Previous Step)Model Access=Direct Target Access, Optimization=Prompt Optimization, History=Previous Step2026.04 | 25 | |
| Agentic Adversarial Rewriting (Attacker Only)Model Access=Direct Target Access, Optimization=None2026.04 | 20.11 | |
| CLAREModel Access=RoBERTa Surrogate2026.04 | 3.9 | |
| TextFoolerModel Access=RoBERTa Surrogate2026.04 | 3.09 | |
| TextBuggerModel Access=RoBERTa Surrogate2026.04 | 2.8 | |
| DeepWordBugModel Access=RoBERTa Surrogate2026.04 | 1.97 |