How about rephrasing that, to not anthropomorphize AI by giving it agency, intent, interests, thoughts, or feelings, and to assign the blame where it belongs:

"If we grant these systems too much power, we could do ourselves serious harm."