Work

QR-Code Phishing Detection (Quishing): A Systematic Review of Attack Vectors, Detection Taxonomies, and Defense Paradigms

Year: 2026
Type: article
Study design: Systematic Review
Abstract: Quick Response (QR) code phishing—widely designated as "quishing" or "QRishing"—has escalated into a severe cybersecurity threat by exploiting the inherent visual opacity of two-dimensional matrix symbologies. Because QR barcodes encode arbitrary alphanumeric strings into geometric arrangements of dark and light modules, human users cannot inspect destination Uniform Resource Locators (URLs) prior to execution, and standard enterprise Secure Email Gateways (SEGs) frequently fail to parse embedded graphical payloads. Consequently, adversaries routinely bypass perimeter filters to execute credential harvesting, malware dissemination, and session hijacking. This systematic literature review synthesizes the state of the art in quishing threat mechanics, benchmark datasets, algorithmic countermeasures, and usable security paradigms. Adhering to the PRISMA 2020 framework, our multi-stage identification screened 1,146 candidate records across eight primary digital repositories (ACM Digital Library, IEEE Xplore, ScienceDirect, SpringerLink, Scopus, Clarivate Web of Science, Nature/Frontiers Open Access, and arXiv/preprints) alongside the MITRE ATT&CK Mobile knowledge base. Following rigorous deduplication and eligibility assessments against predefined inclusion and exclusion criteria, 10 primary empirical research studies were selected for deep comparative synthesis, contextualized against 18 foundational surveys and web-phishing reviews. We establish an updated five-pillar taxonomy: (1) Pre-Scan QR-Image Analysis, (2) Cryptographic Provenance and Blockchain Verification, (3) URL-Based Lexical Machine Learning, (4) Deep Semantic and Multimodal Architectures, and (5) Usable Link-Centric Security Interfaces. Our empirical synthesis resolves previous literature discrepancies by formally decoupling QR-level accuracy (evaluating the complete scan-to-decode pipeline under optical noise, physical blur, and geometric distortion) from URL-level accuracy (classifying isolated decoded string features). Furthermore, we reconcile conflicting reports regarding the SEQR usable scanner, clarifying that its 94.18% metric reflects controlled human user threat identification empowered by visual friction and domain-kerning tooltips, rather than autonomous algorithmic classification. We audit reported performance across all pillars, highlighting that while deep multimodal architectures achieve peak URL-level discriminative accuracy (up to 99.22%), pre-scan pixel classifiers operate without URL decoding (up to 0.9133 AUC), and cryptographic frameworks provide deterministic verification (127.1 ms operational latency) at the expense of closed-ecosystem deployability. Finally, we articulate eight open research challenges—including the scarcity of physically tampered benchmark datasets, cloaking evasion, and mobile inference constraints—and outline concrete trajectories toward adversarially robust, privacy-preserving, and human-resilient quishing countermeasures.
Language: English
Open Access status: green
OpenAlex ID: W7225914737
FWCI:
Cites:
Cited by:
Related to: 10