AI Turned Crypto Fraud Into a $30 Billion Assembly Line

Crypto scam losses hit an estimated $30 billion in 2025, according to TRM Labs. That figure did not come from more criminals working harder. It came from AI running at machine scale, automating fraud that once required a person at a keyboard for every conversation.

TRM Labs tracks illicit cryptocurrency flows globally. Their 2025 report puts total illicit crypto volume at a record $158 billion, a 145% year-over-year increase. AI-enabled scam activity alone grew roughly 500% in a single year. The share of crypto scam reports involving AI elements such as deepfakes and automated chatbots grew about 13 times since 2022.

How the Machine Works

The fraud patterns TRM Labs describes follow a few recurring forms. Generative AI produces realistic profile photos, backstories, and documents. A synthetic "financial advisor" complete with a LinkedIn profile, a headshot, and three years of fake posts feels credible by design.

Voice cloning now requires as little as 20 to 30 seconds of audio. The Arup case showed what comes next: faces that look right, voices that sound right, a video call where nothing is real and $25 million is gone by the time the employee realizes it.

Automated chatbots extend the damage further. Where a human scammer could manage a handful of targets at once, an AI chatbot runs thousands of conversations at the same time, adjusting its pitch in real time. Once money moves, AI-powered laundering tools route it through complex transaction chains with minimal oversight, making recovery nearly impossible.

Every one of these scams works the same way: convincing you that you know and trust the person or institution reaching out. Human detection accuracy for AI-generated voice and video hovers around 55 to 60%, barely better than a coin flip.

The One Thing AI Cannot Fake

AI voice cloning can copy how someone sounds. It cannot know a secret.

Trust Onion gives people a simple way to verify anyone claiming to be someone they know, a bank, or another trusted institution. Three rotating codewords change every 60 seconds, calculated locally on each authorized device with no server required. When someone calls claiming to be a person or organization you trust, one question settles it: "What are the words?"

A cloned voice does not know the words. An automated call does not know the words. They rotate on a schedule, so even if someone overheard them once, they expire within a minute. Note that a failed callback on its own does not prove the original caller was legitimate; the codeword challenge is the check that goes beyond a familiar voice, a spoofed caller ID, or a convincing message.

TRM Labs data shows the tools scammers use are getting better and cheaper, and the gap between what AI can fake and what humans can detect keeps growing. The defense does not have to be technical. It just has to be shared before the call comes.

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.