In March 2023, Jennifer DeStefano received a call from what she believed was her teenage daughter's voice, crying and saying "Mom! I messed up!" A male voice then told her he had kidnapped her daughter and demanded a ransom. DeStefano's daughter was safe — on a ski trip — and the voice was entirely AI-generated. The case, one of the first publicly documented instances of AI voice cloning used in a virtual kidnapping scam, later led DeStefano to testify before the United States Senate.

The call

DeStefano was in a dance studio with her younger daughter when her phone rang. The caller ID showed her older daughter Brianna's number. The voice on the line was indistinguishable from Brianna's — the same cadence, the same intonation, the same emotional register. DeStefano told interviewers she had no doubt it was her daughter.

Seconds later, a man's voice came on the line. He claimed to have Brianna. He demanded USD 1 million, later dropping to USD 50,000. He instructed DeStefano not to hang up.

Other parents in the studio helped her call Brianna's phone on a separate device. Brianna picked up. She was fine. She was in Utah. The call that had sounded unmistakably like her had been fabricated entirely from synthesised audio.

How the technology works

AI voice cloning systems can generate a convincing facsimile of a person's voice from as little as a few seconds of audio. As of 2023, publicly available tools could produce this output in real time, allowing criminals to respond dynamically during a live call rather than playing pre-recorded clips.

The source material is typically gathered from publicly available content — social media videos, YouTube clips, TikToks, voicemails — that the target or their family members have posted online. A teenager who posts regularly on social media may have provided hundreds of hours of voice data without any awareness that it could be misused.

The FBI issued guidance in 2023 noting that virtual kidnapping scams using AI-generated voices were increasing in frequency, particularly targeting parents and grandparents. Callers typically use high-pressure tactics — insisting the call stay open, claiming any verification attempt will endanger the victim — to prevent targets from independently confirming their family member's safety.

Senate testimony

DeStefano testified before the US Senate Special Committee on Aging in July 2023, describing the call in detail. She told senators that the experience was indistinguishable from receiving a genuine distress call from her daughter.

"I had no way of knowing that this wasn't Brianna," she testified. "The voice was perfect."

Her testimony was used to support proposed legislation aimed at requiring disclosure when AI-generated voice content is used in commercial contexts, and to push for stronger penalties for AI-assisted fraud.

Why this is agent crime

The DeStefano case is structurally different from earlier virtual kidnapping scams, in which callers used scripted lines and hoped victims would fill in emotional detail themselves. The AI voice model acted as a synthetic agent — it produced real-time speech that mimicked a specific individual without requiring human impersonation.

The crime did not require a skilled voice actor, a long preparation time, or physical access to the victim. It required audio samples sourced from public platforms and a real-time synthesis tool. Both were commercially available in 2023.

As synthesis technology improves and the barrier to creating convincing voice clones continues to fall, this category of crime is expected to scale. The DeStefano case was widely reported. Most cases are not.