“It is currently unknown if it is possible to keep AI alignment perpetually ahead of AI capabilities or to find a "once-and-for-all" solution to the alignment problem.”