Is Wireless Security Ready for the AI Era? Evidence from a Curated Corpus of 130 Studies, 2020–2026

Abstract

Machine learning (ML) is now central to published radio-link authentication research — and simultaneously the adversary’s standard toolkit. This scoping review classifies 130 curated studies (2020–2026): a 121-study core on ML for wireless-link security or attacks on it (84 defense, 30 threat, 7 dual attack-plus-defense), plus 9 post- quantum-cryptography studies in a separate stratum. Each study is coded on six fields — security function, AI method, track, data source, validation practice, and readiness (R1–R5) — by a single coder and independently validated by a second author. Validation practice splits by kind: 16 of 30 threat studies demonstrate on testbeds or over theair, while simulation-only and standard-split evaluations carry much of the defense mass; closed-loop evaluation appears in 14 of 121 core studies (11.6%). Reproducibility is near- absent (4 verified artifact releases, 3.3%), and standards alignment outside the post-quantum stratum is mention-level only (9.9%). Real-time, update, and multi-agent-trust requirements are essentially unaddressed (1–2 papers each). Because 110 of 130 rows were coded from abstracts by a single coder, cross-group comparisons are indicative rather than measured effects. The paper releases the classified dataset for independent checking, distills six structural research gaps, and closes with a twelve-item reporting checklist (WIRELESS-AI-RC)