De-risking authoritarian AI: a balanced approach to protecting our digital ecosystems