Artificial Intelligence as a Service (AIaaS) offers pre-built AI models to customers looking to leverage AI capabilities with minimal cost and risk. This thesis is predicated on the fact that various pre-built models are being offered as a service, on demand, often with little oversight. Given the generic functionality of the AI models available, and the lack of oversight from service providers, there is a large scope for misuse. This thesis implements and evaluates a variety of mechanisms for uncovering possible AIaaS misuse that can identify ‘signals’ and ‘signatures’ in customer usage data that could indicate potential bad behaviour.
This is my fourth year (masters) university dissertation and the write-up can be found at ACS_diss.pdf
If you have any questions feel free to message on sumaiyah.kola@outlook.com