Towards Trustworthy Agentic AI: Comprehensive Survey Covering Safety, Robustness, Privacy, and System Security

arXiv·medium signal

A comprehensive survey (arXiv:2605.23989, May 23) examines trustworthy agentic AI through two core dimensions critical for high-risk deployments: Safety and Robustness (covering behavioral alignment, adversarial resistance, and failure modes) and Privacy and System Security (covering data protection, access control, and infrastructure hardening). The paper synthesizes the fragmented landscape of agent security research into a unified framework, providing the most complete taxonomy of agent trust requirements published this year. For teams building production agents, this is a checklist — if your system doesn't address both dimensions, it has gaps.