Voice cloning

Voice cloning uses artificial intelligence to generate speech that imitates the vocal characteristics of a particular person using examples of that person's voice.

Recognizing a voice is not verification

A cloned voice can make new speech resemble a known speaker. It differs from simply replaying an existing recording, and it is more specific than generating a generic synthetic voice. Its use to fabricate someone’s speech is a form of audio deepfake.

The FTC describes criminals using cloned voices in family-emergency scams. The same problem applies at work: a message sounding like your manager does not prove they sent it or approved the action.

In May 2025, the FBI reported a campaign using AI-generated voice messages to impersonate senior US officials. The report describes synthetic audio without identifying the generation method behind every message.

Verify the request beyond the audio

Contact the person using details you already trust, rather than details supplied in the suspicious message. If the request concerns a sensitive action, such as releasing information or changing payment details, follow the approval procedure for that action as well.

The FBI cautions that AI-generated content can be difficult to identify. Do not build an approval process that relies on everyone spotting a fake voice. Odd sound quality is not proof of cloning either.

The familiar-voice guide explains how to verify a requester and their authority to approve an action, even when the voice is genuine.

Sources

See it in practice