SafeAgent: A Runtime Protection Architecture for Agentic Systems
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866913045468413952 |
|---|---|
| author | Liu, Hailin Ilyushin, Eugene Ni, Jie Zhu, Min |
| author_facet | Liu, Hailin Ilyushin, Eugene Ni, Jie Zhu, Min |
| contents | Large language model (LLM) agents are vulnerable to prompt-injection attacks that propagate through multi-step workflows, tool interactions, and persistent context, making input-output filtering alone insufficient for reliable protection. This paper presents SafeAgent, a runtime security architecture that treats agent safety as a stateful decision problem over evolving interaction trajectories. The proposed design separates execution governance from semantic risk reasoning through two coordinated components: a runtime controller that mediates actions around the agent loop and a context-aware decision core that operates over persistent session state. The core is formalized as a context-aware advanced machine intelligence and instantiated through operators for risk encoding, utility-cost evaluation, consequence modeling, policy arbitration, and state synchronization. Experiments on Agent Security Bench (ASB) and InjecAgent show that SafeAgent consistently improves robustness over baseline and text-level guardrail methods while maintaining competitive benign-task performance. Ablation studies further show that recovery confidence and policy weighting determine distinct safety-utility operating points. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_17562 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SafeAgent: A Runtime Protection Architecture for Agentic Systems Liu, Hailin Ilyushin, Eugene Ni, Jie Zhu, Min Artificial Intelligence Multiagent Systems Large language model (LLM) agents are vulnerable to prompt-injection attacks that propagate through multi-step workflows, tool interactions, and persistent context, making input-output filtering alone insufficient for reliable protection. This paper presents SafeAgent, a runtime security architecture that treats agent safety as a stateful decision problem over evolving interaction trajectories. The proposed design separates execution governance from semantic risk reasoning through two coordinated components: a runtime controller that mediates actions around the agent loop and a context-aware decision core that operates over persistent session state. The core is formalized as a context-aware advanced machine intelligence and instantiated through operators for risk encoding, utility-cost evaluation, consequence modeling, policy arbitration, and state synchronization. Experiments on Agent Security Bench (ASB) and InjecAgent show that SafeAgent consistently improves robustness over baseline and text-level guardrail methods while maintaining competitive benign-task performance. Ablation studies further show that recovery confidence and policy weighting determine distinct safety-utility operating points. |
| title | SafeAgent: A Runtime Protection Architecture for Agentic Systems |
| topic | Artificial Intelligence Multiagent Systems |
| url | https://arxiv.org/abs/2604.17562 |