An Australian company is using artificial intelligence to make scammers waste time on non-existent victims. The company Apate has created robots that communicate with criminals via phone and the internet, collecting data on fraud and hindering new attempts at deception.
This technology is also used in digital traps designed to attract hackers. The goal is to leverage AI's ability to mimic human behavior to enhance the effectiveness of these defensive strategies.
Apate has been developing this platform for two years. According to Dali Kaafar, the company's founder and CEO, the system includes about 350,000 robots and is utilized by banks with the support of telecommunications companies. These programs answer calls, process text messages, and participate in online groups used for fraudulent activities. During communication, they can gather information about criminal activities.
The company claims to have collected over 250,000 real-time records, including addresses of scam websites, bank details, and accounts used for illegal fund transfers. To avoid suspicion, the robots adopt various profiles and behavioral models; some use WhatsApp, while others do not. There are robots that answer calls, and there are ones that end the conversation, stating they will follow up later.
Dali Kaafar told Wired: 'We like to think we are creating perfect victims for scammers.'
According to him, every minute a criminal spends talking to a robot can prevent them from contacting hundreds or even thousands of people with the same scammer who might use automated dialing systems.
The platform's functionality was tested by journalists from the website Kernel Panic, who tried to convince an artificial victim to invest in cryptocurrency. Despite the conversation proceeding naturally, the attempt failed: after six minutes, the system continued to reject the offer.
Kaafar noted that some calls last more than two hours. While scammers try to convince a non-existent victim, they stop contacting other people.
However, the reach of this strategy has limitations. Criminals make billions of calls and send billions of messages annually, and some operations take place in large physical centers. The robots' activities combine various functions. The same logic applies to so-called 'honeypots'—systems that mimic real computers and services to attract attackers and monitor their methods. Researchers are integrating large language models (LLMs) into these traps to make them more convincing.
In a recent study, Dr. Mark Vero from ETH Zurich and his colleagues found that traps equipped with LLMs held AI agents used in attacks longer than systems with predictable behavior. Vero stated: 'Attacking agents are much more convinced by LLM-created honeypot simulations and identify them as traps much less often.'
In the researcher's opinion, well-designed systems can give an advantage to those involved in defense.
Combating digital crime also depends on cooperation between law enforcement agencies, banks, telecommunications companies, and social networks. Information sharing can help identify patterns and stop operations, but AI is not yet capable of completely preventing scammers from trying to deceive new victims.