analiza aiqbench
This paper addresses a critical vulnerability in LLM security by introducing seven cross-domain detection techniques to overcome the limitations of standard regex and fine-tuned classifier approaches. The proposed methods, including local-sequence alignment from bioinformatics and forensic linguistics, demonstrate significant improvements in F1 scores while managing false positive rates across multiple attack datasets. By porting mechanisms outside traditional AI safety domains, this work offers a robust new paradigm for defending against sophisticated prompt injection attacks.
Beyond Pattern Matching: Seven Cross-Domain Techniques for Prompt Injection Detection
arXiv:2604.18248v3 Announce Type: replace-cross Abstract: Current open-source prompt-injection detectors converge on two architectural choices: regular-expression pattern matching and fine-tuned transformer classifiers. Both share failure modes that recent work has made concrete. Regular expressions miss paraphrased attacks. Fine-tuned classifiers are vulnerable to adaptive adversaries: a 2025 NAACL Findings study reported that eight published indirect-injection defenses were bypassed with greater than fifty percent attack success rates. This work proposes seven detection techniques that each port a mechanism from a discipline outside LLM security: forensic linguistics, materials-science fatigue analysis, deception technology, local-sequence alignment from bioinformatics, mechanism design, spectral signal analysis, and taint tracking. Three are implemented in the prompt-shield v0.4.1 release (Apache 2.0) and evaluated across six datasets (deepset/prompt-injections, NotInject, LLMail-Inject, AgentHarm, AgentDojo). The local-alignment detector lifts F1 on deepset from 0.033 to 0.378 with zero additional false positives on deepset's fifty-six benign samples; the same detector adds ten false positives on the 339-sample NotInject benign set (FPR 0.9%->3.8%), a specificity regression discussed in Section 5.4. The stylometric detector adds 11.1 percentage points of F1 on an indirect-injection benchmark. The fatigue tracker is validated via a probing-campaign integration test. Section 5.6 reports independent evaluation on three peer-reviewed benchmarks (Liu, USENIX Sec 2024; Garak, 2024; InjecAgent, ACL Findings 2024) totalling 8,276 attacks, using the regex-only baseline (d022, d027, d028 disabled). All code, data, and reproduction scripts are released under Apache 2.0.
Abstract:Current open-source prompt-injection detectors converge on two architectural choices: regular-expression pattern matching and fine-tuned transformer classifiers. Both share failure modes that recent work has made concrete. Regular expressions miss paraphrased attacks. Fine-tuned classifiers are vulnerable to adaptive adversaries: a 2025 NAACL Findings study reported that eight published indirect-injection defenses were bypassed with greater than fifty percent attack success rates. This work proposes seven detection techniques that each port a mechanism from a discipline outside LLM security: forensic linguistics, materials-science fatigue analysis, deception technology, local-sequence alignment from bioinformatics, mechanism design, spectral signal analysis, and taint tracking. Three are implemented in the prompt-shield v0.4.1 release (Apache 2.0) and evaluated across six datasets (deepset/prompt-injections, NotInject, LLMail-Inject, AgentHarm, AgentDojo). The local-alignment detector lifts F1 on deepset from 0.033 to 0.378 with zero additional false positives on deepset's fifty-six benign samples; the same detector adds ten false positives on the 339-sample NotInject benign set (FPR 0.9%->3.8%), a specificity regression discussed in Section 5.4. The stylometric detector adds 11.1 percentage points of F1 on an indirect-injection benchmark. The fatigue tracker is validated via a probing-campaign integration test. Section 5.6 reports independent evaluation on three peer-reviewed benchmarks (Liu, USENIX Sec 2024; Garak, 2024; InjecAgent, ACL Findings 2024) totalling 8,276 attacks, using the regex-only baseline (d022, d027, d028 disabled). All code, data, and reproduction scripts are released under Apache 2.0.
v3.0.2 (28 Jul 2026): Abstract scope-hedges specificity to deepset's 56 benign samples and discloses 10 FPs on NotInject (FPR 0.9%->3.8%; Sec. 5.4). Sec. 5.6 reframed as regex-baseline evaluation (d022/d027/d028 disabled in Table 2). Fixes garbled venue for Zhan et al. (NAACL 2025 Findings, arXiv:2503.00061); minor citation cleanup
Cryptography and Security (cs.CR); Computation and Language (cs.CL)
arXiv:2604.18248 [cs.CR]
(or arXiv:2604.18248v3 [cs.CR] for this version)
https://doi.org/10.5281/zenodo.19644135 DOI(s) linking to related resources
Text integral sub licența deschisă a publicației · 291 cuvinte
Subiecte conexe
Game Design Creativity evals
LessWrong · 12 oct. 2026
[AINews] not much happened today
Latent Space · 3 oct. 2026
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
METR · 26 aug. 2026
Closing the Binary Blind Spot: Domain-Knowledge Injection for LLM Malware Triage in npm Packages
Databricks · 16 oct. 2026