Semantic Denial of Service in LLM-controlled robots

This feed aggregates multiple new security-relevant arXiv papers (Apr 29 2026) covering vulnerabilities, threat taxonomies, defenses, and protocols across AI, cryptography, and networking. Key findings: (1) “Semantic Denial of Service in LLM-controlled robots” demonstrates short safety-plausible audio injections can halt or disrupt robot agents (semantic DoS) and shows prompt-level defenses only change disruption modes; unauthenticated audio→LLM pipelines create an avoidable availability dependency. (2) V.O.I.C.E presents an empirically grounded taxonomy of synthetic-voice risks (privacy, mis/