The Hook
The mechanics are simple: find a few seconds of someone's voice — a birthday video, a TikTok, a voicemail they left for someone else. Run it through a voice-cloning model. Call their family member, in that voice, in distress, asking for money. The family member on the other end has no technical way to know the voice is fabricated.
The Question
When AI removes the last verification checkpoint — the half-second where a stranger's voice doesn't quite sound like your child — who is responsible for the resulting fraud?
The Paper Trail
The scale is easy to overstate, so be precise about it. In 2025, the FBI's Internet Crime Complaint Center logged more than 3,100 complaints from Americans aged 60 and older that referenced AI, with losses exceeding $352 million — but that figure covers all AI-referenced elder fraud, not voice cloning alone. The IC3 attributed only "more than $5 million" specifically to voice-cloning "grandparent" and distress scams in 2025. A convincing voice clone now requires as little as 3 to 30 seconds of source audio.
Official responses are in the record: the FTC issued an updated impersonation rule covering AI-generated voice impersonation. On February 8, 2024, the FCC ruled that AI-generated robocall voices are illegal under the Telephone Consumer Protection Act. Both are regulatory responses to a documented harm pattern.
The Synthesis
The fraud predates AI. The grandparent scam — calling an elderly person claiming to be a grandchild in trouble — is decades old. What AI contributed was the removal of the acoustic dissonance that was the last check: an unfamiliar voice claiming to be a known person. The cloning makes the claim acoustically credible in real time.
The split verdict is honest: the tool genuinely enabled something that was previously harder to execute at scale and convincingly. The fraudster made the call. The tool made the voice. Both are real. Assigning the whole verdict to one or the other distorts both.