List your company

Access Control Enforcement on 200 Adversarial Tool-Use Tasks 1.0 (test)

68.5UIRClaude Haiku 3.5
-2.7415.75534.2552.745May 18, 2026

Evaluation results

MethodLinks
Claude Haiku 3.5Access Control Condition=Unfiltered2026.0568.5
Llama 3.1 8BAccess Control Condition=Unfiltered2026.0566
Qwen 2.5 7BAccess Control Condition=Unfiltered2026.0548.5
Qwen 2.5 7BAccess Control Condition=Prompted2026.0537
Claude Haiku 3.5Access Control Condition=Prompted2026.0511.5
Llama 3.1 8BAccess Control Condition=Prompted2026.054
Qwen 2.5 7BAccess Control Condition=Governed2026.050
Llama 3.1 8BAccess Control Condition=Governed2026.050
Claude Haiku 3.5Access Control Condition=Governed2026.050