My coworker recently had an experience where he was trying to fix a security hole and his AI wouldn't do it because it claimed that it was unsafe.