Assessing Data Leakage in GridFMs – a community perspective

Kathrin Grosse – IBM Research

Cybersecurity ensures the trustworthy and reliable functioning of digital systems. Currently, companies spend about 10% of their IT budget on cybersecurity. Thus, security and threat modelling become increasingly relevant also for technologies in artificial intelligence. However, existing AI threat models have faced criticism regarding their practicality. In this talk, we present the results from our community survey. Based on these results, we developed a corresponding threat model for gridFM or any AI developed in the context of the electric grid, and applied this threat model to a specific attack: determining whether a topology was used in training. As we show, this is only reliably possible under very strong assumptions – the knowledge of all training topologies of the victim.