Insurance particularly vulnerable to AI threats, claims new research