A short sample of someone's voice is enough to generate a phone call that sounds like them. The technology is widely available and almost free. A familiar voice is no longer proof of identity, and the kinds of requests that exploit this fact (a relative in trouble, an executive approving an urgent transfer, an assistant confirming a routine instruction) are now in active use.

The samples are not hard to find. A podcast appearance, a conference recording, a brief video on a public social account: each is sufficient. A person whose voice has ever been published is a potential subject. A person whose voice has not been published, but who works in a field where they speak to others, can often be sampled from a recorded call.

What this does to the trust that families and offices place in voice calls is significant. The phone call is the medium where most consequential confirmations have always been made. It is also now the medium most easily forged.

The defence is to agree, in advance, on a way to confirm anything of consequence. A code word known only to family. A second channel for confirming any payment. A pause before any action that cannot be reversed. None of this requires technical skill. All of it requires the discipline to use the agreed channel even when the call sounds genuine.