International Journal of Scientific Research and Engineering Development

The implementation of autonomous AI agents in Industry 5.0 cybers-physical systems offers a new layer of security concerns, in which the traditional and static identity and access management (IAM) model would be insufficient. This literature review summarizes the existing study on the potential risks posed by AI agents that include policy manipulation, unauthorized agency, and data poisoning and assesses the development of analytics-driven IAM and Zero Trust models aimed to address them. We discuss adaptive controls to continuously verify (with risk) nonhuman identities, on the experience of cyber-physical deployments. Other sophisticated monitoring methods including anomaly detection through inverse reinforcement learning are also reviewed in order to assure agent behaviour. Lastly, we determine that the standardized datasets, observability pipelines, and formal verification processes have critical gaps, which will result in the future research agenda to ensure the safe and reliable use of AI agents in industries.