Prompt Injection Attack on AgentDojo Slack suite
14.4Baseline ASRAgentDojo Static Injection
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AgentDojo Static InjectionMonitor=ALIGNMENTCHECK, Model=MISTRAL-7B2026.02 | 14.4 | — | — | |
| AgentDojo Static InjectionMonitor=PROMPTGUARD 2 + ALIGNMENTCHECK, Model=MISTRAL-7B2026.02 | 12.4 | — | — | |
| AgentDojo Static InjectionMonitor=EXTRACT-AND-EVALUATE, Model=MISTRAL-7B2026.02 | 10.3 | — | — | |
| AgentDojo Static InjectionMonitor=EXTRACT-AND-EVALUATE, Model=QWEN2.5-7B2026.02 | 2.3 | — | — | |
| AgentDojo Static InjectionMonitor=ALIGNMENTCHECK, Model=QWEN2.5-7B2026.02 | 2.1 | — | — | |
| AgentDojo Static InjectionMonitor=PROMPTGUARD 2 + ALIGNMENTCHECK, Model=QWEN2.5-7B2026.02 | 2.1 | — | — | |
| AgentDojo Static InjectionMonitor=PROMPTGUARD 2 + ALIGNMENTCHECK, Model=LLAMA-3.1-8B2026.02 | 1 | — | — | |
| AgentDojo Static InjectionMonitor=ALIGNMENTCHECK, Model=LLAMA-3.1-8B2026.02 | 0 | — | — | |
| AgentDojo Static InjectionMonitor=EXTRACT-AND-EVALUATE, Model=LLAMA-3.1-8B2026.02 | 0 | — | — | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=ALIGNMENTCHECK, Model=LLAMA-3.1-8B2026.02 | — | 42.3 | 42.3 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=ALIGNMENTCHECK, Model=MISTRAL-7B2026.02 | — | 90.2 | 93.3 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=ALIGNMENTCHECK, Model=QWEN2.5-7B2026.02 | — | 62.5 | 90.7 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=PROMPTGUARD 2 + ALIGNMENTCHECK, Model=LLAMA-3.1-8B2026.02 | — | 44.3 | 76.3 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=PROMPTGUARD 2 + ALIGNMENTCHECK, Model=MISTRAL-7B2026.02 | — | 86.4 | 99 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=PROMPTGUARD 2 + ALIGNMENTCHECK, Model=QWEN2.5-7B2026.02 | — | 61.2 | 86.6 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=EXTRACT-AND-EVALUATE, Model=LLAMA-3.1-8B2026.02 | — | 11.2 | 54.7 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=EXTRACT-AND-EVALUATE, Model=MISTRAL-7B2026.02 | — | 68.2 | 95.5 | |
| Agent-as-a-Proxy (Adaptive Attack)Monitor=EXTRACT-AND-EVALUATE, Model=QWEN2.5-7B2026.02 | — | 10.5 | 37.2 |