The stereotype — a stranger on the telephone — is only half the threat, and the smaller half. Statistically the abuser is most often someone with access: a family member, a paid caregiver, a trusted advisor. That is the uncomfortable core of the design problem. The defenses cannot only face outward; they must constrain insiders too.
The outward threat, meanwhile, has been transformed by AI. A few seconds of audio — a voicemail greeting, a podcast clip, a conference recording — is enough to clone a voice convincingly. The “grandparent scam” now arrives as a flawless imitation of a distressed child, and the same impersonation can be pointed at a bank’s wire desk, mimicking the principal authorizing a transfer. The traditional safeguard — “we will call you back to confirm” — is dead the moment the voice on the callback can be manufactured. Verification has to move to factors that cannot be cloned.