Australia's 350,000 AI Scambaiters Turn the Tables on Fraudsters

A company in Australia has built an unusual workforce: 350,000 artificial-intelligence agents whose only job is to waste the time of scammers. The firm, Apate.ai, takes its name from the Greek goddess of deception, and its bots are designed to be deceived only in appearance. Each agent poses as an ordinary person — young or old, male or female — with different accents, several languages and its own personality, according to a report in the Chinese science outlet Guokr.

The target is not the public but the fraud industry itself. When a scam call comes in, or a phishing site asks for details, an AI persona answers. It chats, hesitates, asks for clarification, forgets a password, misreads a number. The aim is to keep the fraudster occupied for as long as possible, which costs the scam operation time and money while giving real people a quieter phone line. The same approach extends to investment fraud and fake websites, where bots can fill in forms with plausible but useless information.

Scambaiting is not new. For years, hobbyists have answered fraudulent emails with elaborate stories and fake identities, sharing the results online. What is different here is scale. A single human can handle one or two conversations at a time; a network of hundreds of thousands of software agents can handle many at once, around the clock, in several languages. The company says the personas vary enough that a fraudster cannot easily tell one from another.

The idea sits at an awkward intersection of security and ethics. Defenders of the approach argue that scammers are criminals who profit from fear and confusion, and that slowing them down protects victims. Critics might ask whether automated deception, even against wrongdoers, risks normalising the practice, or whether a bot that impersonates a vulnerable older person could draw retaliation against real people. The report does not settle those questions, and the technology is still young.

For ordinary users, the practical lesson is simpler. The best defence against a scam call remains the same: hang up, do not share codes or bank details, and report the number. An AI decoy does not remove the need for that caution. But it does add a new layer to an old arms race — one in which the fraudsters now have to wonder whether the voice on the other end is a person at all.

What makes the story notable is the economics. Fraud operations rely on volume; if enough calls are answered by bots that never pay, the cost of each successful scam rises. That is the same logic behind spam filters and call-blocking apps, applied to conversation rather than delivery. Whether it works at scale will depend on how quickly scammers learn to detect synthetic voices.

For global readers, the case is a reminder that AI’s first mass-market role may not be writing essays or generating pictures but sitting in the middle of ordinary transactions, including the dishonest ones. The next few years will show whether automated decoys become a standard tool for banks and telecoms, or remain a clever experiment run by one company with a classical name.