Reporting on Google GTIG and Mandiant’s October 30 analysis describes AI penetration-testing tools such as ARTEX being abused for internal reconnaissance, permission analysis, and subsequent activity in real attacks. One record contained 27 activities in 23 seconds, not a measurement of bypassing every external security barrier. Separately checked Wiz research documents repeated attempts against authentication and account-recovery paths at software vendors, many of which failed. Speed should not be confused with flawless autonomous compromise; defenses need least privilege, short-lived credentials, restricted management interfaces, and prompt containment of abnormal behavior.