# ATBench: A Diverse and Realistic Agent Trajectory Benchmark for Safety Evaluation and Diagnosis

arXiv: https://arxiv.org/abs/2604.02022 · HTML: https://mlsanity.com/paper/2604.02022/

## Key facts

- Cited by 36 papers in the map (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 19 of its 29 references that are in the map.
- First posted on arXiv 2 Apr 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)

## Builds on (most cited first, 19 of 19)

- [ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs](https://mlsanity.com/paper/2307.16789/) (2023)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [ToolAlpaca: Generalized Tool Learning for Language Models with 3000 Simulated Cases](https://mlsanity.com/paper/2306.05301/) (2023)
- [Agent-SafetyBench: Evaluating the Safety of LLM Agents](https://mlsanity.com/paper/2412.14470/) (2024)
- [R-Judge: Benchmarking Safety Risk Awareness for LLM Agents](https://mlsanity.com/paper/2401.10019/) (2024)
- [AgentSpec: Customizable Runtime Enforcement for Safe and Reliable LLM Agents](https://mlsanity.com/paper/2503.18666/) (2025)
- [ToolACE: Winning the Points of LLM Function Calling](https://mlsanity.com/paper/2409.00920/) (2024)
- [Qwen3Guard Technical Report](https://mlsanity.com/paper/2510.14276/) (2025)
- [ShieldAgent: Shielding Agents via Verifiable Safety Policy Reasoning](https://mlsanity.com/paper/2503.22738/) (2025)
- [OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents](https://mlsanity.com/paper/2506.14866/) (2025)
- [AgentAuditor: Human-Level Safety and Security Evaluation for LLM Agents](https://mlsanity.com/paper/2506.00641/) (2025)
- [AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security](https://mlsanity.com/paper/2601.18491/) (2026)
- [Building a Foundational Guardrail for General Agentic Systems via Synthetic Data](https://mlsanity.com/paper/2510.09781/) (2025)
- [ProbGuard: Proactive Runtime Monitoring for LLM Agent Safety via Probabilistic Prediction](https://mlsanity.com/paper/2508.00500/) (2025)
- [SafeToolBench: Pioneering a Prospective Benchmark to Evaluating Tool Utilization Safety in LLMs](https://mlsanity.com/paper/2509.07315/) (2025)
- [EgoNight: Towards Egocentric Vision Understanding at Night with a Challenging Benchmark](https://mlsanity.com/paper/2510.06218/) (2025)
- [Are Your Agents Upward Deceivers?](https://mlsanity.com/paper/2512.04864/) (2025)
- [Domain-RAG: Retrieval-Guided Compositional Image Generation for Cross-Domain Few-Shot Object Detection](https://mlsanity.com/paper/2506.05872/) (2025)
- [Poly-Guard: Massive Multi-Domain Safety Policy-Grounded Guardrail Dataset](https://mlsanity.com/paper/2506.19054/) (2025)

## Built on by (most cited first, 30 of 36)

- [OSWorld 2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks](https://mlsanity.com/paper/2606.29537/) (2026)
- [Auditing Agent Harness Safety](https://mlsanity.com/paper/2605.14271/) (2026)
- [PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents](https://mlsanity.com/paper/2608.04003/) (2026)
- [TRACES: Proactive Safety Auditing for Multi-Turn LLM Agents via Trajectory-State Modeling](https://mlsanity.com/paper/2605.27690/) (2026)
- [AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security](https://mlsanity.com/paper/2605.29801/) (2026)
- [Safety Testing LLM Agents at Scale: From Risk Discovery to Evidence-Grounded Verification](https://mlsanity.com/paper/2607.01793/) (2026)
- [SHE: Trajectory-driven Safety Harness Evolution for LLM Agents](https://mlsanity.com/paper/2608.09885/) (2026)
- [StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing](https://mlsanity.com/paper/2608.24777/) (2026)
- [Long-Horizon Agent Trajectory Attribution: A Unified Benchmark and Fine-Grained Annotation Framework](https://mlsanity.com/paper/2608.06909/) (2026)
- [OpenSkillRisk: Benchmarking Agent Safety When Using Real-World Risky Third-Party Skills](https://mlsanity.com/paper/2607.20121/) (2026)
- [AgentS4D: Benchmarking Runtime Risks across the Execution Lifecycle of LLM-Based Workspace Agents](https://mlsanity.com/paper/2607.27294/) (2026)
- [Automata from Agent Traces: Failure and Next-Step Prediction](https://mlsanity.com/paper/2608.23670/) (2026)
- [RUBAS: Rubric-Based Reinforcement Learning for Agent Safety](https://mlsanity.com/paper/2606.04051/) (2026)
- [Agent Hacks Agent: Autoresearch for Production-Agent Red-Teaming](https://mlsanity.com/paper/2607.11698/) (2026)
- [SingGuard-NSFA: Extensible Guardrails for Agentic AI via Generative Reasoning and Real-Time Classification](https://mlsanity.com/paper/2607.13081/) (2026)
- [AgentDrift: A Step-Labeled Benchmark of Injection-Hijacked LLM Agent Trajectories](https://mlsanity.com/paper/2609.06972/) (2026)
- [Content-Aware Attack Detection in LLM Agent Tool-Call Traffic: An Empirical Study of Features, Architectures, and Evaluation Protocols](https://mlsanity.com/paper/2605.11053/) (2026)
- [PRISM: Preference-Aware Influence Function Based Data Selection Method for Efficient Fine-Tuning](https://mlsanity.com/paper/2605.21422/) (2026)
- [Online Agent-as-a-Judge: Situation-Generating Evaluation for Interactive Agents](https://mlsanity.com/paper/2606.08200/) (2026)
- [ClawTrack: Towards Trace-Level Evaluation and Improvement of Real-World Autonomous Agents](https://mlsanity.com/paper/2607.28037/) (2026)
- [RePolicy: Reinforcement Learning for Safety-Policy Invocation in Agent Safeguards](https://mlsanity.com/paper/2608.24275/) (2026)
- [AURA-Eval: Evaluation Framework for Acting Under Risk Awareness in LLM Agent Trajectories](https://mlsanity.com/paper/2609.06783/) (2026)
- [Skynet: Workflow-Level Anomaly Detection for Agentic AI via Semantic and Structural Modeling](https://mlsanity.com/paper/2609.06835/) (2026)
- [DriftNet: A Dual-Head Trajectory Transformer for Detecting and Localizing Prompt Injection in LLM Agents](https://mlsanity.com/paper/2609.10892/) (2026)
- [HazardAuditor: From Executable Threats to Safer Computer-Use Agents](https://mlsanity.com/paper/2609.15134/) (2026)
- ["You're Right, Let Me Fix It": How LLM Agents Damage Correct Work When Falsely Accused](https://mlsanity.com/paper/2609.32616/) (2026)
- [Trust the Brand, Lose Control: How Identity Hijacks LLM Agent Orchestration](https://mlsanity.com/paper/2609.32635/) (2026)
- [Agent Safety From Within: Detecting Harmful Trajectories from LLM Internal States](https://mlsanity.com/paper/2609.33039/) (2026)
- [Auditing Agent Actions through Query-Conditioned Attribution](https://mlsanity.com/paper/2609.33676/) (2026)
- [Deny Without Disabling: Authorization-Paired Evaluation and Control for Multi-Agent Systems](https://mlsanity.com/paper/2610.00371/) (2026)

## Abstract (arXiv)

Evaluating the safety of LLM-based agents is increasingly important because risks in realistic deployments often emerge over multi-step interactions rather than isolated prompts or final responses. Existing trajectory-level benchmarks remain limited by insufficient interaction diversity, coarse observability of safety failures, and weak long-horizon realism. We introduce ATBench, a trajectory-level benchmark for structured, diverse, and realistic evaluation of agent safety. ATBench organizes agentic risk along three dimensions: risk source, failure mode, and real-world harm. Based on this taxonomy, we construct trajectories with heterogeneous tool pools and a long-context delayed-trigger protocol that captures realistic risk emergence across multiple stages. The benchmark contains 1,000 trajectories (503 safe and 497 unsafe), averaging 9.01 turns and 3.95k tokens, with 1,954 invoked tools drawn from pools spanning 2,084 available tools. Data quality is supported by rule-based and LLM-based filtering plus full human audit. Experiments on frontier LLMs, open-source models, and specialized guard systems show that ATBench is challenging even for strong evaluators, while enabling taxonomy-stratified analysis, cross-benchmark comparison, and diagnosis of long-horizon failure patterns.