Researchers at the University of Surrey have developed software that can verify how much information an AI data system has learned from an organization’s digital database, making it an essential part of a company’s online security protocol. The software can also detect if AI has identified and can exploit flaws in software code. The research won the best paper award at the 25th International Symposium on Formal Methods.
The verification software can identify how much AI can learn from their interactions and whether they have too much knowledge, which could break privacy, making it possible for organizations to deploy AI safely in secure settings. Professor Adrian Hilton, Director of the Institute for People-Centred AI at the University of Surrey, believes that tools that can verify the performance of generative AI models are essential to their safe and responsible deployment.
