Post-Hoc Trajectory-Risk Certification for Modular LLM-Based Security Agents

Security-focused arXiv feed containing research on LLM-agent assurance and prompt-injection defenses, adversarial degradation of vision-transformer efficiency, public-safety radio signaling confidentiality, multi-hop microservice access control, blockchain aggregation integrity, automotive ECU firmware patch analysis, and AI-agent security classification. The most directly actionable findings concern passive inference and possible unencrypted voice extraction in TETRAPOL, undocumented vulnerabilities and residual flaws in a vehicle Brake ECU patch, and denial-of-service-style compute inflation