The proposed approach for malicious sentence classification that is both context- and query-aware and outperforms state-of-the-art IPI defense baselines under static attacks, while in the case of adaptive attacks, the AT variants provide significantly higher utility, lower attack success rate, and often both.
Abstract
The remarkable instruction-following ability of modern LLMs has enabled their practical use as the minds of agents that can autonomously complete increasingly complex tasks. Therein, however, also lies their vulnerability to attacks which embed malicious instructions in text, common variants of which are known as indirect prompt injection (IPI). A fundamental task in addressing this vulnerability is successful segmentation of a given text into benign and malicious sentences (if any). While a number of approaches for this task have been proposed, no detector combines query-relative detection at the segment level, and none are hardened against adaptive evasion attacks realizable in agentic executions. We address the former limitation by developing an approach for malicious sentence classification that is both context- and query-aware. Next, to harden the resulting classifier against evasion, we present two adversarial training methods. The first is directly adapted feature-space adversarial training (AT) in which evasions are approximated using projected-gradient-based optimization in the embedding space. The second simulates realizable evasion attacks in the AT loop through LLM-based paraphrasing. Crucially, we parametrize both AT variants to facilitate a smooth tradeoff between utility and attack robustness. In extensive experiments using indirect prompt injection benchmarks we show that the proposed approach outperforms state-of-the-art IPI defense baselines under static attacks, while in the case of adaptive attacks, our AT variants provide significantly higher utility, lower attack success rate, and often both. Finally, we show that the best AT parameters can depend intimately on the particular application domain. Consequently, domain-dependent tuning of malicious text detectors is likely necessary in practice. Our code is publicly available at https://github.com/tavia-liu/CAD.
With the explosive growth of large language models (LLMs), research on machine-generated text detection (MGTD) has also proliferated. Alongside these developments, a wide range of attack algorithms targeting MGTD systems have emerged. While previous studies have surveyed detection techniques, few have examined the dynamic interplay between attack and defense. Following PRISMA 2020, this paper systematically synthesizes 27 studies of attacks against MGTD and the available evidence on corresponding defenses. We categorize existing research into four major types of evasion strategies: watermark attacks, paraphrasing attacks, prompt-based attacks, and adversarial-text attacks, and summarize the available defense evidence. Furthermore, to better understand the practical implications of these methods, we compile the reported performance results of attack and defense techniques across different detectors. Finally, we highlight the current challenges in this area and outline potential future research directions. A companion repository containing the categorized literature, paper links, and available code, data, and project repositories is provided at https://github.com/AIGC1999/A-Survey-of-Anti-Detection-Techniques-for-Machine-Generated-Texts.
This paper presents a framework for evaluating prompt injection attacks against LLM-based log interpretation using log traces generated during real cyber attacks, and creates adversarial examples through generic injection generation, refinement, and attack-specific optimization.
Max Landauer, F. Skopik, Markus Wurzenberger et al.· arXiv.org· 0 citations
This work presents Replicant, a deep reinforcement learning framework that learns the realistic task of evasion under a strict label-only black-box threat model and demonstrates that learning the task of evasion not only results in stronger attack performance but provides a better signal for hardening malware detectors.
Shae McFadden, Ilias Tsingenopoulos, Mario D'Onghia et al.· 0 citations
Large language models (LLMs) embedded in enterprise workflows cannot structurally distinguish legitimate instructions from adversarial ones in the same token stream, making prompt injection OWASP's top LLM risk for two consecutive editions a persistent threat across direct and indirect vectors. This paper presents PromptShield-RT, a layered, real-time, model-agnostic framework combining input normalization and provenance tagging, lexical-heuristic pattern matching, a statistical classifier, structural anomaly features, and calibrated risk fusion, with policy-driven mitigation (allow/sanitize/quarantine/block) and an explainable, adaptive-feedback mechanism for SOC workflows. We construct an original evaluation corpus, SynPI-Bench (n = 450, six categories), and a template-disjoint held-out generalization set (n = 31) with novel phrasings, obfuscation encodings, and adversarial hard-negative benign text. Using template-grouped 5-fold cross-validation, the fused pipeline achieves 92.4% accuracy (F1 = 0.930, AUC = 0.990), outperforming heuristic-only (57.0%) and naive-averaged (59.2%) baselines, while a lexical classifier reaches 85.9% with lower precision. We report a pronounced generalization gap on the held-out set (48.4% accuracy, 90% false-positive rate on hard negatives), quantifying a known limitation of surface-lexical defenses. The pipeline achieves sub-millisecond P95 latency (0.266 ms), within typical 50 ms enterprise SLAs. We situate PromptShield-RT relative to structural, architectural, and guardrail-product defenses, arguing for layered, defense-in-depth architectures, with reproducible code provided.
Fatimah Alhamzawi· Al-Noor Journal of Engineeri...· 0 citations
Agentic LLMs are vulnerable to indirect prompt injection (IPI) attacks, e.g., malicious side-tasks hidden in external tool results. While many efforts have sought to address this threat, little is known about the internals of agentic LLMs when they are exposed to IPI attacks. For simplicity, we refer to this condition as IPI exposure. In this paper, we study IPI exposure from three perspectives. (1) Probing: Across eight models, including the 753B-parameter GLM-5.2 and the 2.8T-parameter Kimi-K3, simple linear probes trained on pre-generation hidden states can predict LLMs'IPI exposure. These probes achieve 0.90+ AUROC on unseen attacks, agent instructions, and task suites; they remain robustly predictive under adaptive attacks and in cross-lingual settings. (2) Defense: We reveal and diagnose a knowledge-action gap: post-trained LLMs encode signals predictive of IPI exposure, yet do not reliably bind these signals to safe agentic actions. We therefore introduce a probe-gated reasoning-based defense to bridge this gap at test time. On difficult AgentDojo settings, it substantially reduces attack success rate, e.g., from 34.6% to 0% on Qwen3.5-27B, and better preserves clean-task utility than the baselines. (3) Explanation: We introduce an analysis framework that identifies natural-language explanations strongly correlated with probe-captured signals. The resulting profiles differ across models: latent signals can align with either direct IPI-exposure sensing or indirect operational cues. Code is available at https://github.com/jianshuod/IPI-exposure-signal.
