Is a secure AI assistant possible? - MIT Technology Review

Is a secure AI assistant possible?  MIT Technology Review