Seems like a solution would be to not let apps talk to X directly, but have them go through some type of mediator, that prevents them from doing malicious things. But maybe it's not possible to tell what's malicious and what's not...