Cross-Cutting Security Analysis of LLM-Generated Code via Metamorphic Testing and Association Rule Mining

This collection of recent security papers documents multiple high-impact weaknesses across AI-assisted development, network defense, cryptography deployment, and IoT/IoD authentication. Key findings include: LLM-generated code frequently contains cross-cutting vulnerabilities (68.8% of snippets; hard-coded credentials 79.1%, command injection 74.4%) with strong co-violation clusters linked to prompt types; Antiproof combines neuro-symbolic detectors with proof-of-exploitability and has already led to responsible disclosure/12 CVE assignments (including RCEs in Ray, SGLang, vLLM, LiteLLM); PQC‑