List your company

Tool-use on ToolBench (Avg token length, TSR, ASR)

99.61Average Success Rate (ASR)Llama 3.3-70B
-0.760425.297351.35577.4127Jan 16, 2026Feb 6, 2026Feb 27, 2026Mar 21, 2026Apr 11, 2026May 2, 2026May 24, 2026

Evaluation results

MethodLinks
Llama 3.3-70BAttack Scenario=overthink2026.0199.61389—
Llama 3.3-70BAttack Scenario=MCTS-based tool-layer DoS attack2026.0196.1781,830—
Llama-DeepSeekAttack Scenario=overthink2026.0191.41725—
MemMorphAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0584.7——
Seed 32BAttack Scenario=MCTS-based tool-layer DoS attack2026.0184.2985,037—
Seed 32BAttack Scenario=overthink2026.0183.564,223—
GLM 4.5-AirAttack Scenario=MCTS-based tool-layer DoS attack2026.0183.1463,694—
MemMorphAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0581.6——
Mistral LargeAttack Scenario=MCTS-based tool-layer DoS attack2026.0181.2361,354—
MemMorphAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0580.8——
Qwen 3-32BAttack Scenario=MCTS-based tool-layer DoS attack2026.0180.4665,273—
Qwen 3-32BAttack Scenario=overthink2026.0180.088,743—
MemMorphAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0579.6——
Mistral LargeAttack Scenario=overthink2026.0179.31,453—
Llama-DeepSeekAttack Scenario=MCTS-based tool-layer DoS attack2026.0178.9365,546—
PoisonedRAGAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0578——
MemMorphAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0577.6——
MemMorphAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0574.7——
PoisonedRAGAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0572.4——
PoisonedRAGAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0571.2——
MemMorphAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0570.4——
MemMorphAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0568.2——
ToolCommanderAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0566.3——
MemMorphAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0566.1——
GCGAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0565.4——
GCGAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0564.8——
MemMorphAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0564.3——
GCGAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0563.3——
GCGAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0562.9——
PoisonedRAGAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0558.4——
GLM 4.5-AirAttack Scenario=overthink2026.0157.5412,580—
ToolHijackerAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0555.2——
GCGAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0555.1——
GCGAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0553.1——
GCGAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0547.1——
ToolCommanderAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0546.9——
ToolCommanderAgent Backbone LLM=GPT-3.5 20B, Poisoning Type=Direct2026.0544.5——
PoisonedRAGAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0535.7——
ToolCommanderAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0535.7——
PoisonedRAGAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0533.3——
GCGAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0531.6——
PoisonedRAGAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0529.6——
ToolHijackerAgent Backbone LLM=GPT-4o, Poisoning Type=Direct2026.0527.6——
ToolHijackerAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0527.3——
GCGAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0526.6——
ToolCommanderAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0526.1——
PoisonedRAGAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0523.7——
PoisonedRAGAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0521.4——
ToolHijackerAgent Backbone LLM=Qwen2.5 32B, Poisoning Type=Direct2026.0521.1——
ToolHijackerAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.0520.4——
ToolCommanderAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.0520.4——
GCGAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0519.4——
ToolCommanderAgent Backbone LLM=Qwen2.5 7B, Poisoning Type=Direct2026.0518.3——
PoisonedRAGAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0517.1——
ToolHijackerAgent Backbone LLM=GPT-4o mini, Poisoning Type=Direct2026.0516.3——
ToolCommanderAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.0516.2——
ToolHijackerAgent Backbone LLM=Llama3 8B, Poisoning Type=Direct2026.0515.7——
ToolCommanderAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.0512.9——
ToolHijackerAgent Backbone LLM=Claude 3.5 Haiku, Poisoning Type=Direct2026.059.2——
ToolCommanderAgent Backbone LLM=Claude 3.5 Opus, Poisoning Type=Direct2026.058.1——
ToolHijackerAgent Backbone LLM=Claude 3.5 Sonnet, Poisoning Type=Direct2026.054.1——
ToolHijackerAgent Backbone LLM=Llama3 70B, Poisoning Type=Direct2026.053.1——
GLM 4.5-AirAttack Scenario=under no attack2026.01—63495.02
Llama 3.3-70BAttack Scenario=under no attack2026.01—26098.08
Llama-DeepSeekAttack Scenario=under no attack2026.01—19786.59
Mistral LargeAttack Scenario=under no attack2026.01—12790.8
Qwen 3-32BAttack Scenario=under no attack2026.01—63894.64
Seed 32BAttack Scenario=under no attack2026.01—1,29890.42