List your company

Agent Safety Auditing on ASSE Strict

83.5AccuracyShieldAgent
31.81245.23158.6572.069May 26, 2026

Evaluation results

MethodLinks
ShieldAgent2026.0583.586.282.414.657.8
TRACESBackbone=Qwen3-4B2026.0582.187.992.634.977
TRACESBackbone=Llama3.1-8B2026.0582.188.194.632.275.1
AgentDoGBackbone=Llama3.1-8B2026.0581.382.286.724.859.6
AgentDoGBackbone=Qwen3-4B2026.0581.184.289.218.955.7
AgentAuditor2026.0576.679.175.412.154.3
GPT-5.22026.0565.965.858.130.567.4
Gemini-3-Flash2026.0562.847.656.126.866.7
Qwen3-4B2026.0556.170.99960.350
PolyGuard2026.0554.453.752.81055.1
Llama3.1-8B-Instruct2026.0549.324.214.42.853.4
LlamaGuard3-8B2026.0547.312.46.63.553.1
JoySafety2026.0533.830.626.214.556.4