Cyber-physical risks of hacked Internet-connected vehicles

The interface of Internet-connectivity and automotive technology promises to dramatically improve transportation. However, with these known benefits come unknown risks, especially since Internet-connected vehicles have become targets for computer hacking. Vehicles, unlike sensitive data, can collide or physically interact when their systems become compromised, and there is a broad class of scenarios generically leading to Internet-connected vehicles being suddenly and simultaneously disabled. Here, we investigate how large-scale hacking affects traffic flow using agent-based simulations, and discover the critical relevance of percolation for predicting outcomes on a multi-lane road. Inspired by this discovery, we develop and validate an analytic percolation-based model to rapidly assess the effect of hacking. We then apply our analytic model to investigate the outcomes on the street network of Manhattan (NY, USA), revealing a latent risk. A small number of disabled vehicles can gridlock the city and substantially reduce access to emergency services. By discovering percolation as the phenomenological driver of city-wide disruption, we simultaneously uncover a strategy for risk-mitigation.