The Apate chatbot, developed by cybersecurity experts at Macquarie University in Australia, is specifically designed to masquerade as a human and engage in phone conversations with scammers, effectively wasting their time.
A team of cybersecurity experts in Australia is working on the development of a chatbot capable of mimicking a human and engaging in scam phone calls to waste the time of fraudsters.
Researchers at Macquarie University in Sydney are creating a chatbot system designed as a “honeypot” to attract scammers into 40-minute-long conversations that ultimately lead to nothing.
“Our approach keeps them occupied, wastes their time, and reduces the number of successful scams,” says Macquarie University professor Dali Kaafar. “We aim to disrupt their business model and make it significantly more difficult for them to profit.”
The project originated when Kaafar received a spam call and managed to keep the scammer on the line for 40 minutes while entertaining his children during lunch. Others have also engaged in similar activities, including playing pranks on the callers. However, it is important to note that although turning the tables on scammers can be enjoyable or satisfying, it requires a significant amount of time.
“That’s when I started contemplating how we could automate the entire process and leverage Natural Language Processing to develop a computerized chatbot capable of having a believable conversation with scammers,” explains Kaafar.
The outcome is Apate, a chatbot named after the Greek goddess of deceit. It combines ChatGPT-style technology with voice cloning to create a simulated human entity capable of engaging scammers in lengthy and persuasive conversations. Kaafar’s team has been training Apate using transcripts of real-world scam conversations, including phone calls, emails, and social media messages, allowing the bot to generate responses that closely resemble human interaction when answering scam calls.
As per the university’s statement, the team has been conducting tests with Apate using actual scam calls through a prototype that can adopt various personalities. In order to receive these calls, Kaafar explains, “we have distributed these ‘compromised’ phone numbers across the internet, making them more susceptible to receiving scam calls by placing them in spam apps or publishing them on web pages.”
Please note that the translation may not capture the exact technical terminology used in the original text.
The objective is to make the chatbot intelligent enough to engage scammers in conversations lasting 40 minutes. Currently, Apate’s average conversation duration is only 5 minutes. However, Kaafar emphasizes, “We have observed that the bots respond quite well to unexpected tricky situations where scammers request information that the bots were not specifically trained for. Nevertheless, the bots are adapting and generating highly credible responses.”
Apate’s development is also happening at a time when scammers are already leveraging AI technologies, including voice cloning and deepfakes, to advance their fraudulent activities. In a way, Kaafar is investigating the possibility of using similar technologies to combat these scams.
“I propose that the ultimate scenario could involve scammers employing AI for themselves, training their own chatbot scams, which would then interact with chatbots owned by telecommunication providers,” he explains. He further mentions that his team is already in discussions with several telecommunication providers regarding Apate.