AI Bots Waste Scammers’ Time and Gather Fraud Intel

AI “victims” tie up scammers

For two years, the Australian company Apate has built a system that routes phone scammers to AI bots trained to keep talking without falling for the scam. Founder and CEO Dali Kaafar says the goal is to give scammers enough hope to stay on the line while the platform collects intelligence about the fraud.

Apate says its platform is used by banks and supported by telecom companies, with around 350,000 bots. The bots answer calls, join scam chat groups, and reply to text messages. The company says it has gathered more than 250,000 pieces of real time fraud intelligence, including scam URLs, money mule accounts, and bank details.

Testing the perfect victim

Apate says the bots have different personalities, language skills, and profiles so scammers are less likely to detect AI. Some personas have WhatsApp and some do not. Some pick up the phone and some hang up and say they will call back later.

In a demo where WIRED played the scammer, the AI personas were skeptical but left openings to keep the conversation going. The timing felt natural during a test that included a friend named Lucy and a financial adviser named Mickey. After six minutes, the testers still had not convinced the AI to invest in their cryptocurrency opportunity. Kaafar says Apate calls regularly last more than two hours.

Honeypots get more realistic

Cybercriminals still send billions of messages and calls each year, and some operations run industrial scale scamming sites. Researchers say AI could support a new era of anti scam work through better intelligence sharing among police, social media companies, banks, and other industries, plus AI driven monitoring and data analysis.

Defenders are also using false virtual machines called honeypots to attract hackers, waste their time, and learn their techniques. Mark Vero, a doctoral researcher at ETH Zurich, says open source honeypot providers are adding large language models to make systems look more realistic. In recent research, Vero and colleagues found that an LLM powered honeypot kept AI agents attacking significantly longer and was flagged as a honeypot at a much lower rate.

Sources: Wired Ai