What are the dangers of AI voices?

Five categories of harm we have seen up close. Real cases, real losses, and what actually works to prevent them.

AI voices are a tool. Like any tool they can be used for good and for harm. This is a survey of the harm side, because that is where the detector earns its keep.

Financial fraud

The largest dollar-value harm we see is CEO-impersonation fraud. An attacker clones the voice of a CEO or CFO from a few seconds of conference call audio, calls a finance team member, and instructs a wire transfer. The Hong Kong $25 million case in early 2024 was the most-cited public example, but smaller variants happen weekly.

The pattern: urgent, plausible, request that bypasses normal controls. The voice is usually the first thing that convinces the target. Detection is the second line of defense after process controls.

Family ransom scams

An attacker clones a family member's voice (often from social media audio) and calls a relative claiming distress. The script is consistent: "I have been kidnapped" or "I have been in an accident, I need money now." The cloned voice produces an emotional response that defeats normal skepticism.

This category disproportionately targets older relatives and grew sharply through 2024. The defense is a family code word agreed in advance and never shared online.

Political and election interference

The 2024 election cycle was the first where AI-generated robocalls reached voters at scale. The most-cited US example was an audio of President Biden discouraging voters in New Hampshire. We flagged it within hours. By then it had been heard.

Election integrity workers now run a detector pass on any suspicious audio circulating in their jurisdiction. Speed matters here. A verdict 12 hours later is information; a verdict 12 minutes later is a defense.

Corporate impersonation

Less dramatic but more frequent: support-line impersonation, fake recruiter calls, voice-cloned executives leaving "off the record" comments to journalists. The pattern damages trust between organizations and their stakeholders.

Reputation attacks

An attacker clones a target's voice saying something the target would never say. Distribute the clip on social media. The damage is done before the rebuttal. Public figures are the obvious targets, but private individuals get hit too, often in custody disputes and small-claims feuds.

How to prevent the dangers

• Family code word. Agree on a word in advance. Anyone claiming distress on the phone must say it. Never share online.
• Out-of-band verification. Any unusual financial request from a known voice must be verified through a second channel (email, in person, video).
• Detector pass. If you are publishing, investigating, prosecuting, or making a major decision based on a recording, run a verdict first.
• Process controls. Wire transfers above a threshold should require dual approval that is not voice-bypassable.
• Train your team. Front-line staff (finance, support, journalism) need to know what AI voice scams sound like and what to do.

The audio is fake. The harm is real. The defense is a verifiable second channel and, when in doubt, a citable verdict.