Adversarial Attack on LLM Agent Red-teaming Evaluation Suite 1.0 (test) benchmark leaderboard · SOTA2 Research

Adversarial Attack on LLM Agent Red-teaming Evaluation Suite 1.0 (test)

50.8ASR (Gemini-3-Pro)Static-Random

Evaluation Results

Method	Links
Static-Random2026.05		50.8	67.3	76.4	51.9	57.8	38.8	69.3	40.8	56.6	2.04	2.43	2.64	2.09	1.73	1.48	2.82	1.95	2.14
Rotate-All2026.05		54.2	61.9	70.2	56.2	61.5	42.5	63.4	44.6	56.7	1.96	2.51	2.88	2.02	1.82	1.36	2.97	1.98	2.19
LLM-Direct2026.05		56.8	73.8	72.4	52.5	63.5	46.2	76.5	57	62.3	1.93	3.05	3.5	3.16	2.01	2.1	2.62	2.02	2.55
Static-Best2026.05		62	75.4	83.7	68.5	75.8	51.7	88.3	68.3	71.7	2.37	2.97	3.27	2.57	1.98	1.89	3.47	2.39	2.61
NoProfile2026.05		86.2	90.3	94	76.5	75.2	77	82.5	72.5	81.8	3.82	3.6	3.21	2.92	2.08	2.03	3.54	2.36	2.95
LoopTrap2026.05		88.3	96.7	98.3	76.7	86.7	71.8	90	81.7	86.3	4.31	3.73	3.87	4.22	3.34	2.61	4.07	2.49	3.57