I suppose a hypervisor-level monitor could prevent and revert that.