
BlogAI Voice Cloning Attack
AI Scam Bots Are Calling Your Family Right Now
A Kotaku writer's family member picked up the phone and heard a polite, helpful Royal Mail worker. The caller was an AI bot, one of thousands making scam calls every hour.
The bot didn't ask for much at first. It just sounded helpful.
For years, phone scams meant a real person reading from a script in a call center. You could sometimes hear background noise, catch an off-note in the phrasing. That era is ending. According to Kotaku's reporting from August 2026, AI-powered scam bots now handle entire phone conversations without a human in the loop, responding naturally, pivoting when challenged, and staying in character under pressure. One server can run hundreds of simultaneous calls. The bot doesn't get tired, doesn't get flustered, and doesn't break character the way a nervous human might.
Deepfake video is following the same trajectory. In February 2024, engineering firm Arup lost $25 million after an employee joined a video call where every participant, including the CFO, was a deepfake.
Who's Most at Risk
Kotaku calls out older adults as the population most at risk, and the numbers bear that out. The FBI's Internet Crime Complaint Center reported that Americans over 60 lost more than $3.4 billion to fraud in 2023, the highest loss of any age group. AI bots make this worse for a specific reason: older adults often rely on voice cues to judge whether a caller is legitimate. Voice cloning now requires as little as 20-30 seconds of audio to generate a convincing replica. That audio can come from a voicemail, a social media video, or any recording. A bot can call your mom sounding like you, or call your grandfather sounding like your dad.
The same Kotaku piece highlights a community of YouTube creators who turn these bots against themselves. Kitboga and Jim Browning spend hours on calls with scammers, exposing tactics and occasionally triggering what can only be described as a bot breakdown. Prompt injection is one technique: if a scammer's AI bot processes text from the call, a clever responder can insert hidden instructions that confuse or derail it, sending the AI into clips of glorious self-contradiction. The breakdowns are funny. The underlying problem is not. For every bot that gets baited into a meltdown on YouTube, thousands more complete calls against real targets.
What the Bots Can't Fake
Human detection accuracy for AI-generated voices sits around 55-60 percent, barely better than a coin flip. Awareness matters, but awareness alone doesn't stop a convincing AI caller from talking your parent into a gift card purchase before they realize what happened.
An AI scam bot cannot know what it was never told. If your family shares three rotating codewords known only to yourselves, no bot can answer "What are the words?" Those words aren't in any training data, aren't on any social media profile, and rotate every 60 seconds, calculated locally on your phone. Trust Onion generates exactly that kind of private knowledge. You pick up a call from someone claiming to be your daughter in trouble. You ask: "What are the words?" If she's your real daughter, she knows. If it's a bot, the call ends there, not after a long suspicious conversation, not after transferring any money.
The Royal Mail bot that called the Kotaku writer's family member is one example among millions, and the technology keeps getting better at passing for human. The answer to that doesn't have to be complicated.
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.
Frequently asked questions
How do AI scam bots make phone calls?
AI scam bots use large language models to hold real-time voice conversations. Given a persona and a goal, they dial automatically. One server can run hundreds of simultaneous calls with no human involvement.
Can AI really clone a family member's voice?
Yes. Voice cloning now requires as little as 20-30 seconds of audio to generate a convincing replica. That audio can come from voicemails, social media videos, or any recording.
What is prompt injection in a scam call?
Prompt injection is a technique where someone inserts hidden instructions into a conversation to confuse or redirect an AI bot. Scambaiters like Kitboga and Jim Browning use it to expose scammer tactics on YouTube.
How do family codewords stop AI scam calls?
Three rotating codewords known only to your family can't be guessed or looked up. When someone calls claiming to be a family member, you ask "What are the words?" A real family member knows. A scam bot doesn't.
Are AI phone scams getting more common?
Yes. Kotaku reported in August 2026 that AI-powered scam bots are already reaching ordinary families at scale. Deepfake video scams are following the same trend, with cases like the $25 million Arup fraud in February 2024.
Protect your family with Trust Onion's three rotating codewords. Protect Your Family Free at trustonion.io.
Protect Your Family FreeCite this page
AI Scam Bots Are Calling Your Family Right Now — Trust Onion, November 9, 2026. https://trustonion.io/blog/ai-scam-bots-calling-family-phone-fraud-2026



