Security and Assurance of AI Agents in Cyber-Physical and Autonomous Systems
Not stated
- Funding
- Self-Funded PhD Students Only
- Application deadline
- Year-round applications
About the project
About the Project AI agents and large-language-model-based autonomous systems are beginning to influence software engineering, cyber defence, robotics, transport, industrial automation and cyber-physical infrastructure. While these systems offer significant benefits, they also introduce new security risks, including unsafe tool use, prompt injection, data leakage, adversarial manipulation, poor explainability, supply-chain exposure and loss of human oversight. These risks become more serious when AI agents interact with physical systems, operational technology or safety-critical digital infrastructure. This PhD will investigate security and assurance methods for AI agents operating in cyber-physical and autonomous environments. The project will examine how AI agents perceive system state, select actions, interact with tools, respond to adversarial inputs and affect cyber-physical resilience. Research may include threat modelling for AI agents, adversarial testing, secure agent architectures, human-in-the-loop control, explainable decision-making, red-team evaluation, digital-twin-based simulation and assurance metrics for safe deployment. The project is suitable for a candidate interested in AI security, cyber security, autonomous systems, software assurance or cyber-physical systems. The work may involve Python, LLM/agent frameworks, attack simulation, digital twins, security testing and experimental evaluation. The expected outcome is a set of methods and demonstrators for assessing whether AI agents can be trusted in operational and safety-critical environments.