The emerging EU AI Act call for risk assessment to be conducted for AI system in specific high risk applications and also for generative foundational AI systems that could be used in such applications. The AI Act and its supporting technical standards do not however provide guidance on how risk assessments from foundational model vendors ,e.g. [1], could be used to support risk assessment needed for AI application that use those models, e.g. in field such as medical devices [2] or digital forensics.
This project will build on existing work on developing semantic models [3] and explore how LLM risk assessments could be imported, used and expanded to address risks in a specific regulated domain.