Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.07256

Efficient Auditing of Adversarial AI Agent Behavior from Agent Traces

Eugene Zhang·Cheng-Yun King Yang·Dongyan Xu
View original source ↗Read source document ↗Export metadata ↓

Abstract

AI agents powered by large language models (LLMs) can perform complex tasks but may harm the systems they operate in, either intentionally or unintentionally. Existing agent monitoring approaches rely on rule-based guardrails or LLM-based trace auditing. However, rule-based guardrails can be bypassed through obfuscation and may miss harmful actions beyond their predefined rules, whereas applying an LLM to audit every action is costly. We present a two-stage agent trace auditing framework. The first stage uses single-event and trace-sequence rules to select pending actions for inspection; the second uses an LLM audit agent to examine each selected action in the context of the agent's preceding trace before execution. We jointly refine the gate rules and audit instructions using training data, allowing the framework to adapt to complex agent behaviors rather than relying solely on predefined rules. On the public benchmark OpenAgentSafety, our framework reduces the average number of LLM audits from 8.15 to 2.33 per run and token usage from 47.8k to 14.6k, with a detection rate of 72.8\% compared with 81.5\% when every action is audited. In two simulated multi-agent case studies, the framework flags all malicious traces while reducing audit token usage by more than 80\%.

Explore related subjects

cs.CR

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Eugene Zhang, Cheng-Yun King Yang, Dongyan Xu. 2026-10-05. Efficient Auditing of Adversarial AI Agent Behavior from Agent Traces. https://arxiv.org/abs/2610.07256

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

A Survey on Reconfigurable Intelligent Surfaces Towards Practical Deployments: Security and Privacy Perspectives

Reconfigurable Intelligent Surfaces (RIS) have emerged as a transformative technology capable of reshaping wireless environments through dynamic manipulation of electromagnetic waves. Although RISs have been widely studied for communication, sensing, and localization, their security and privacy implications are still fragmented across application-specific studies, prototypes, and testbeds. This survey provides a deployment-oriented review of RIS applications, with a focus on the security and privacy threats, vulnerabilities, and defensive strategies relevant to practical use. We analyze scenarios with two types of systems (with and without legitimate RIS) and two types of attackers (with and without malicious RIS), and clarify how RIS introduces new attack surfaces to practical systems, including eavesdropping, jamming, and spoofing attacks. In response, we review defenses against RIS-related attacks in these systems, such as protocol- and algorithm-level protections, unauthorized RIS detection, and interference/disruption-based mitigation. We also discuss scenarios in which the legitimate user applies an additional RIS to defend against attacks. To support future research, we also curate papers and available open resources at https://awesome-ris-security.github.io/. By highlighting RIS's functionality and its security/privacy challenges and opportunities, this survey aims to guide researchers and engineers toward the development of secure, resilient, and privacy-preserving RIS-enabled practical wireless systems and environments.

cs.CR↗
arXiv2026

Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code

Large Language Models (LLMs) are increasingly used for code generation, raising concerns that they may be misused to produce malicious code. Meanwhile, Grammar-Constrained Decoding (GCD) has been widely adopted to improve the reliability of LLM-generated code by enforcing syntactic validity. In this paper, we reveal a counterintuitive risk: this reliability-oriented technique can itself become an attack surface. We uncover a new jailbreak attack, termed CodeSpear, that exploits GCD to induce LLMs into generating malicious code. Our experiments show that simply applying a benign code grammar constraint can effectively jailbreak LLMs. To address this vulnerability, we propose CodeShield, a safety alignment approach that robustly preserves safe behavior even under attacker-controlled grammar constraints. CodeShield aligns the model in the code modality by teaching it to generate honeypot code under GCD. Such code is semantically harmless, so it does not implement the malicious request, and structurally diverse, so it is difficult to suppress through grammar tightening. At the same time, CodeShield still preserves natural-language refusals when natural language is available. Experiments on 10 popular LLMs across 4 benchmarks show that CodeSpear outperforms representative jailbreak baselines and increases the attack success rate by more than 30 percentage points on average. CodeShield also restores safety under CodeSpear while preserving benign utility. Our findings reveal a fundamental risk of GCD and call for greater attention to its potential security implications.

cs.CR↗
arXiv2026

Jaguar: Fast Private Neural Network Inference with Power-of-Two Homomorphic Arithmetic

Hybrid HE/2PC private neural network inference remains bottlenecked by prime-modulus homomorphic arithmetic in convolution and by a precision flow that runs ReLU at doubled bitwidth before invoking a separate truncation protocol. We present Jaguar, a system built on a single design choice--a power-of-two ciphertext ring--that addresses both. The choice enables SPA-Conv, a coefficient-domain convolution kernel that replaces NTT-centric polynomial multiplication with Scalar-Polynomial Accumulation, and an exact ciphertext-side truncation by local right shifts that lets ReLU run directly at the target fixed-point precision and eliminates the post-ReLU truncation protocol. Where NTT remains genuinely useful--at the client, for the single polynomial multiplication during decryption--we recover it through an auxiliary NTT prime, preserving the power-of-two protocol substrate while keeping decryption O(N log N). On ImageNet-scale ResNet-18, ResNet-50, and MobileNetV2 with AVX disabled, Jaguar achieves 2.07-3.72x lower end-to-end latency than Cheetah and 2.16-3.36x lower than Rhombus, with 1.16-1.76x lower communication than Cheetah.

cs.CR↗