Supervisors

Suitable for

Abstract

As generative AI systems are increasingly widely used in a growing range of applications, understanding the potential for cyber-attacks to compromise the integrity and confidentiality of the data they output is critical. This project would aim to develop attack graphs for generative AI systems, and based on this design cybersecurity test suites that facilitate testing the security of generate AI systems against a range of potential attacks. The project might involve using these test suites to test the security of a range of generative AI implementations in a laboratory setting.