What Guides the Agent? Adjudicating Unauthorized Behavior via Localizing Behavior-Guiding Instructions

• URL: http://arxiv.org/abs/2608.24022v1
• Date: Tue, 25 Aug 2026 03:24:53 GMT
• Title: What Guides the Agent? Adjudicating Unauthorized Behavior via Localizing Behavior-Guiding Instructions
• Abstract summary: We propose Attnlocate, a runtime framework for fine-grained localization of context spans that genuinely influence tool-calling decisions.<n>Attnlocate casts this localization problem as an object detection task, aiming to detect the distinctive activation traces induced by behavior-guiding instructions.<n>We evaluate Attnlocate across ten agent configurations from five LLM families, covering scenarios involving indirect prompt injection and tool poisoning.
• Score: 16.444770499696457
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: LLM agents integrated with external resources gain complex task capabilities, yet the unified natural-language context channel makes them vulnerable to injection attacks: untrusted external data may be dynamically parsed as behavior-guiding instructions during LLM inference, thereby subverting the agent's decision. Existing defenses focus on static detection or isolation of malicious content at the input/output level, remains insufficient for detecting such dynamic inducements that arise during model reasoning. We propose Attnlocate, a runtime framework for fine-grained localization of context spans that genuinely influence tool-calling decisions, i.e., behavior-guiding instructions. Attnlocate casts this localization problem as an object detection task, aiming to detect the distinctive activation traces induced by behavior-guiding instructions within the attention matrix. Specifically, we design a multi-head, multi-layer attention aggregation scheme to construct a token-level feature space tailored for object detection. Then, a 1-D U-Net equipped with an anchor-free detection head is deployed to detect these spans. Finally, based on the authority of the provider from which the detected behavior-guiding spans originate, Attnlocate dynamically adjudicates malicious invocation attempts. We evaluate Attnlocate across ten agent configurations from five LLM families, covering scenarios involving indirect prompt injection and tool poisoning. Attnlocate achieves a mean IoU of 0.743, an average AUROC of 0.956, and a 0.934 true-positive rate at 0.067 false-positive rate. It also transfers effectively across unseen models and supports authority policy adaptation without retraining.

Related papers

• ScrambleToolBench: Agents Search Exhaustively Even When Their Own Map Points to the Next Step [74.83535434786145]
  We introduce ScrambleToolBench, an interactive terminal benchmark designed to isolate behavioral reasoning.<n>The benchmark requires agents to uncover hidden tool behaviors entirely through trial-and-error interaction.<n>Our evaluation of state-of-the-art language models reveals that successful initial discovery does not translate into robust adaptation.
  arXiv  Detail & Related papers  (2026-08-03T15:07:46Z)
• Linguistic Firewall: Geometry as Defense in Multi-Agent Systems Routing [1.4033701678475552]
  We introduce ANTAP, an evaluation-driven routing architecture that discards indirect proxies in favor of active capability testing.<n>ANTAP achieves near-zero ASR against description-based injection attacks, compared to 67.3% and above for the description-based router baseline.
  arXiv  Detail & Related papers  (2026-06-29T16:51:06Z)
• Who Pays the Price? Stakeholder-Centric Prompt Injection Benchmarking for Real-world Web Agents [93.19140872946842]
  Web agents driven by large language models (LLMs) are increasingly deployed in real-world environments.<n>This makes them vulnerable to prompt-injection attacks, in which seemingly benign content embeds adversarial instructions that manipulate agent behaviour.<n>We introduce textbfsysname, a benchmark to systematically categorize and attribute harm in real-world web agent systems.
  arXiv  Detail & Related papers  (2026-06-11T14:12:43Z)
• From Knowing to Acting: Benchmarking Self-Awareness Capability of LLM Agents [69.07911499189129]
  We introduce KAPRO, a framework that evaluates cognitive-behavioral alignment by decoupling an agent's metacognitive judgment (Knowing) from its spontaneous execution (Acting)<n>Experiments show that self-awareness capability is strongly correlated with task success but degrades sharply in internal-capability settings.<n> proprietary and reasoning-oriented models demonstrate more reliable cognitive gating.
  arXiv  Detail & Related papers  (2026-06-09T17:17:08Z)
• A Low-Latency Fraud Detection Layer for Detecting Adversarial Interaction Patterns in LLM-Powered Agents [18.524016493711386]
  Large Language Model (LLM)-powered agents demonstrate strong capabilities in autonomous task execution, tool use, and multi-step reasoning.<n> adversarial interactions can manipulate agent behavior through direct prompt injection, indirect content attacks, and multi-turn escalation strategies.<n>We propose a low-latency fraud detection layer for detecting adversarial interaction patterns in LLM-powered agents.
  arXiv  Detail & Related papers  (2026-05-01T22:35:59Z)
• Beyond Input Guardrails: Reconstructing Cross-Agent Semantic Flows for Execution-Aware Attack Detection [32.301679396929536]
  We propose SysName, a framework that shifts the defensive paradigm from static input filtering to execution-aware analysis.<n>SysName synthesizes fragmented operational primitives into contiguous behavioral trajectories, enabling a holistic view of system activity.<n> Empirical evaluations demonstrate that SysName effectively detects over ten distinct compound attack vectors, achieving F1-scores of 85.3% and 66.7% for node-level and path-level end-to-end attack detection, respectively.
  arXiv  Detail & Related papers  (2026-03-04T01:59:16Z)
• Mask-Based Window-Level Insider Threat Detection for Campaign Discovery [0.0]
  We study unsupervised window-level insider threat detection on the CERT r4.2 dataset.<n>We show that explicitly separating activity presence from activity magnitude yields substantial performance gains.<n>Across multiday attack campaigns lasting between one and seven days, the proposed approach achieves a window-level precision-recall AUC of 0.71.
  arXiv  Detail & Related papers  (2026-02-11T16:43:01Z)
• The Why Behind the Action: Unveiling Internal Drivers via Agentic Attribution [63.61358761489141]
  Large Language Model (LLM)-based agents are widely used in real-world applications such as customer service, web navigation, and software engineering.<n>We propose a novel framework for textbfgeneral agentic attribution, designed to identify the internal factors driving agent actions regardless of the task outcome.<n>We validate our framework across a diverse suite of agentic scenarios, including standard tool use and subtle reliability risks like memory-induced bias.
  arXiv  Detail & Related papers  (2026-01-21T15:22:21Z)
• DRIFT: Dynamic Rule-Based Defense with Injection Isolation for Securing LLM Agents [52.92354372596197]
  Large Language Models (LLMs) are increasingly central to agentic systems due to their strong reasoning and planning capabilities.<n>This interaction also introduces the risk of prompt injection attacks, where malicious inputs from external sources can mislead the agent's behavior.<n>We propose a Dynamic Rule-based Isolation Framework for Trustworthy agentic systems, which enforces both control and data-level constraints.
  arXiv  Detail & Related papers  (2025-06-13T05:01:09Z)
• Defending against Indirect Prompt Injection by Instruction Detection [109.30156975159561]
  InstructDetector is a novel detection-based approach that leverages the behavioral states of LLMs to identify potential IPI attacks.<n>InstructDetector achieves a detection accuracy of 99.60% in the in-domain setting and 96.90% in the out-of-domain setting, and reduces the attack success rate to just 0.03% on the BIPIA benchmark.
  arXiv  Detail & Related papers  (2025-05-08T13:04:45Z)
• Object-Centric Latent Action Learning [70.3173534658611]
  We propose a novel object-centric latent action learning framework that centers on objects rather than pixels.<n>We leverage self-supervised object-centric pretraining to disentangle action-related and distracting dynamics.<n>Our results show that object-centric pretraining mitigates the negative effects of distractors by 50%.
  arXiv  Detail & Related papers  (2025-02-13T11:27:05Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.