A team at the University of California, Santa Cruz has published new research showing how visual-language AI models that help control self-driving cars can be exploited or hijacked with carefully ...