Senate Hearing: AI Is Targeting Seniors and Winning

Seniors in the United States lost $7.7 billion to scams in 2025. That figure, cited at the committee's hearing titled 'The AI Deception Machine: Deepfakes, Chatbots, and the New Frontier of Senior Fraud,' is almost certainly low: experts testified that many victims never report what happened. What made July 29 different from past elder-fraud hearings was the technology at the center of it. This is no longer about phishing emails.

Two witnesses gave the numbers a face. The first was that mother. A caller screamed in her daughter's voice, claimed the daughter had been kidnapped, and demanded money. The voice was indistinguishable from her child's. She nearly sent the funds before the call collapsed. Her daughter was fine and had no idea any of it was happening. The second witness was a surgeon whose face and voice had been used in a deepfake video to promote a fraudulent health product, a video he never agreed to, never recorded, spread online under his real name and credentials.

These are not edge cases. The technical skill required to run these scams has collapsed. Voice cloning now requires just 20 to 30 seconds of audio, pulled from a voicemail, a social media video, or a video call recording. Once a scammer has a usable sample, a convincing fake call takes minutes, with no acting and no accent training. Experts before the committee were direct: generative AI is lowering the barrier for criminals faster than law enforcement can respond, and a national anti-scam strategy is no longer optional.

The Grandparent Scam, Upgraded

The kidnapping call is a variation on the grandparent scam. In the classic version, a caller pretends to be a grandchild in trouble, arrested or in an accident, and asks for money fast and quietly. "Don't tell Mom and Dad." That script has worked for years because it exploits love and urgency. Add a cloned voice and the scheme becomes nearly impossible to detect in the moment. You are not hearing a stranger pretend to be your grandchild. You are hearing your grandchild's voice, the pauses, the emotion, the specific way they say your name, all of it generated from audio scraped from a birthday video or a graduation post.

Policy responses matter, and the committee's call for a national anti-scam strategy suggests more coordinated federal action may be coming. But policy moves slowly. Scam calls do not.

A Check That Takes Three Seconds

The most reliable protection available right now is a shared verification system that no AI can fake. Trust Onion gives families three rotating codewords that change every 60 seconds, calculated locally on each person's phone. When a call arrives from someone claiming to be a grandchild or a daughter in danger, one question settles it: "What are the words?"

A scammer working from a cloned audio sample has no way to know codewords that exist only on the authorized devices in your circle and expire before anyone can steal them. Calling the number back or reaching someone at a known number does not prove the original caller was legitimate. The codewords are the check. The caller must still know your family's private three words.

The mother who heard her daughter screaming had no quick way to confirm what she was hearing. The surgeon whose face appeared in a deepfake had no way to prevent it. The gap between what scammers can do today and what most older adults know to watch for is wide, and it grows every time the tools get cheaper.

The Senate heard that story. The question is whether the people in your life will hear it before the next call comes in.

A controlled voice-cloning safety demonstration

Try our tool to see just how easy voice cloning is. This controlled safety demonstration exists only to show how easy voice cloning is and to make the risk tangible. It is not a real incident or proof of anyone's identity. Your submitted sample is handled transiently to generate the fixed demonstration audio. Trust Onion does not identify you or the person from the sample. Trust Onion does not store the submitted recording, clone, or generated audio.