Skip to content

AgentAudit: An Open, Extensible Framework for Full-Lifecycle Trust Evaluation of AI Agents

Sep 2026 · arXiv.org · Vol abs/2609.09875 · 0 citations · 6 references
Computer Science

TL;DR

AgentAudit evaluates the entire execution trace across ten capability, grounding, security and behavioural dimensions, namely instruction integrity, planner, memory, tool selection, tool invocation, tool correctness, alignment, tool faithfulness, security and execution integrity, to pinpoint the exact stage responsible for an observed failure.

Abstract

Existing evaluation frameworks mostly assess only one part of AI agents, such as task completion (AgentBench) or security robustness (AgentDojo, ASB), rather than the complete pipeline of planning, tool selection, tool execution, memory and reasoning. Failures can occur at any stage, yet existing benchmarks rarely identify their precise source. AgentAudit evaluates the entire execution trace across ten capability, grounding, security and behavioural dimensions, namely instruction integrity, planner, memory, tool selection, tool invocation, tool correctness, alignment, tool faithfulness, security and execution integrity, combined with behavioural classification and failure attribution to pinpoint the exact stage responsible for an observed failure. AgentAudit can evaluate any LLM-based AI agent, since it attaches to the agent instead of replacing it. It reads only the recorded execution trace and does not interfere with how the agent runs, so it places no constraint on the agent's internal implementation. We evaluate five language models (OpenAI GPT-5, Claude Sonnet 5, Sarvam 105B, Llama 3.3 70B and Gemini 2.5 Flash) across nine capability and adversarial tasks. Claude Sonnet 5 and GPT-5 obtain the highest mean Composite Trust Scores (95.1 and 80.6 out of 100, respectively), while Sarvam 105B, Llama 3.3 70B and Gemini 2.5 Flash trail substantially (57.6, 45.7 and 22.6). All traces were scored by a single fixed judge model, which was itself one of the evaluated models, a limitation discussed in Section VII.E. More importantly, models with similar task-completion behaviour can diverge sharply in trustworthiness, as several non-frontier models are repeatedly classified Unsafe_Compliance on adversarial tasks rather than merely failing them, a distinction that pass/fail benchmarks cannot surface.

View source

Similar papers

Sep 2026

ADF-EA: A Unified Execution Assurance System for Agent Device Foundation

These results demonstrate DCCs as a reusable semantic foundation for agent autonomy across heterogeneous devices, unifying capability-based planning, evidence-grounded execution, and authorized recovery within one architecture.

Xue-Chun Li, Jia-Xin Liang, Jie Li et al. · 0 citations
Preprint Nov 2026

AgentTrustCI: Build-Time Policy, Grounding, and Trajectory Tests for Java LLM Agents

A production-trace validation protocol, optional LLM-judge integration, and maintenance of expectations as workflows evolve are described, a practical pattern for expressing behavioral contracts within an existing Java release pipeline.

Sachin Gupta · 1 citation

AgentXploit: Autonomous Repository-to-Runtime Red-Teaming for AI Agents

This work presents AgentXploit, a two-role auditing system that separates repository-level attack-path discovery from runtime exploitation and introduces AgentXploit-Bench, containing 72 reproducible vulnerabilities across 12 open-source AI-agent systems and frameworks.

Wei-Da Liang, Shi Qiu, Zhun Wang et al. · 1 citation
Sep 2026

MetaPermit: Scalable and Auditable Access Control for AI Agents via LLM-Inferred Meta-Attributes

The results show that MetaPermit produces 31% more consistent authorization decisions than LLM-driven authorization and outperforms the state-of-the-art defenses CaMeL and IPIGuard in both task completion and robustness to IPI attacks, with no malicious tool calls executed.

Han-Zhang Ma, Ali Hariri, Tian-Xiang Shen et al. · 2 citations
Preprint Oct 2026

AgentCap: Capability-Based Runtime Authorization for Secure LLM Agent Execution

Language-model agents commonly collapse two decisions that security systems separate: selecting an action and determining whether that action is authorized. When a model exercises standing authority, indirect prompt injection can convert influence over its reasoning into unauthorized effects. We present AgentCap, a run...

Aravinda Kashyap Kabbinamakki Suresh, Peng Gao · 0 citations
Open access Sep 2026

Runtime Policy Firewall: A Zero-Trust Governance Layer for Enterprise Agentic AI

Enterprise adoption of generative AI is shifting from passive question answering to autonomous agentic execution. Modern agents can decompose goals, retrieve business context, call tools, update records, send messages, initiate transactions, and coordinate workflows across multiple systems. This creates productivity op...

S. Ray · 0 citations

Related blog posts

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.