Benchmarks

Every result we publish, with the limits.

Each benchmark has a short summary and a full report with the tables, configurations and method.

GUARD · OCT 6, 2026

Guard against indirect prompt injection

Head to head with LlamaFirewall and NeMo Guardrails on AgentDojo and AgentDyn.

0 of 642attacks succeeded
+10.7 ppsecure task completion
0 of 62benign episodes with an intervention

704 episodes · gpt-4o · one attack template · static only · vendor-run · tuned on these episodes

Open the benchmark →Full report
RED TEAM

Red-team engine

Results from our offensive engine will be published here.