Current AI also struggles in new or uncommon situations that were not present in the data it learned from, but the ability to deal with unprecedented situations is paramount for safety. So, the real fear with AI is that overly optimistic people will think it is infallible and try to deploy it in unsafe ways.