Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses

This collection of recent papers surveys and advances security across AI, networking, and authentication domains. Key contributions include: a comprehensive taxonomy and risk analysis of safety, attacks (adversarial, backdoor, jailbreak, hardware) and defenses for embodied AI (highlighting multimodal perception fragility and planning instability); a dynamic jailbreak-detection method (SALO) that raises detection from ~0% to >90% by tracing refusal trajectories; FreeUp, a frequency-decoupled anomaly detector for encrypted network traffic addressing spectral mismatch; Contrastive Privacy, a semn