Banks and telecom companies are using AI-generated victims to keep scammers occupied and collect intelligence about their operations. Australian company Apate says its system now includes about 350,000 personas that answer calls, respond to messages and enter online scam groups.
The bots are designed to sound interested without handing over money or useful personal information. They vary in language, personality and availability so a criminal does not encounter the same predictable response each time. Apate says some scam calls continue for more than two hours, time that cannot be spent contacting real targets.
The conversations also produce evidence. According to the company, its system has gathered more than 250,000 pieces of fraud intelligence, including malicious links, bank details and accounts used to move stolen money. The approach extends older “honeypot” security systems, which imitate vulnerable computers to attract attackers and observe their methods.
Separate research at ETH Zurich found that large-language-model honeypots kept automated attacking agents engaged significantly longer than static decoys and were less often recognized as traps. These systems do not solve the global scam problem: criminals still send billions of calls and messages, and evidence must reach banks, platforms and law enforcement quickly. Their practical value is narrower—raising attackers’ costs while giving defenders more current information about active campaigns.