Paper 2601.18491
AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security
- Published
- Jan 2026
- Research lab
- Independent
- Citations
- 24
- GitHub
- 680 stars
01 In brief
Summary
AgentDoG is a diagnostic guardrail framework for AI agent safety and security, developed by the Shanghai Artificial Intelligence Laboratory.
It addresses limitations in existing guardrails by introducing a unified three-dimensional safety taxonomy that categorizes agentic risks by source (where), failure mode (how), and real-world harm (what).
Guided by this taxonomy, the authors created ATBench, a fine-grained agent safety benchmark with 500 trajectories (250 safe, 250 unsafe), averaging 8.97 turns and covering 1,575 unique tools, using an unseen-tools evaluation setting.
AgentDoG models, available in 4B, 7B, and 8B parameter sizes across Qwen and Llama families, are trained via supervised fine-tuning on over 100k synthesized trajectories.
The framework performs trajectory-level safety evaluation and fine-grained risk diagnosis, and includes an agentic XAI attribution module that traces unsafe actions to specific steps or sentences.
Experimental results show AgentDoG achieves state-of-the-art performance, e.g., 92.8% accuracy on ATBench, outperforming existing guard models and competing with larger general models.
All models and datasets are openly released.
02 From the paper
Abstract
The rise of AI agents introduces complex safety and security challenges arising from autonomous tool use and environmental interactions. Current guardrail models lack agentic risk awareness and transparency in risk diagnosis. To introduce an agentic guardrail that covers complex and numerous risky behaviors, we first propose a unified three-dimensional taxonomy that orthogonally categorizes agentic risks by their source (where), failure mode (how), and consequence (what). Guided by this structured and hierarchical taxonomy, we introduce a new fine-grained agentic safety benchmark (ATBench) and a Diagnostic Guardrail framework for agent safety and security (AgentDoG). AgentDoG provides fine-grained and contextual monitoring across agent trajectories. More Crucially, AgentDoG can diagnose the root causes of unsafe actions and seemingly safe but unreasonable actions, offering provenance and transparency beyond binary labels to facilitate effective agent alignment. AgentDoG variants are available in three sizes (4B, 7B, and 8B parameters) across Qwen and Llama model families. Extensive experimental results demonstrate that AgentDoG achieves state-of-the-art performance in agentic safety moderation in diverse and complex interactive scenarios. All models and datasets are openly released.