Lancaster University researchers, commissioned by the UK’s Department for Science, Innovation and Technology, analyzed more than 9,100 peer-reviewed AI security studies published since 2020 and identified five key research gaps: data integrity, AI infrastructure security, end-user risk, data and model provenance, and safe model disposal. The review found that agentic AI systems represent an emerging threat area with particularly limited security research. The authors call for more holistic, lifecycle-wide approaches to AI security research to keep pace with agentic system deployment.
