Creating Fake Victims Using Artificial Intelligence to Disrupt Criminal Business Models

2023-06-27

Macquarie University cybersecurity experts have developed a multilingual chatbot aimed at diverting scammers' time towards making fake phone calls, ultimately reducing the number of people worldwide who suffer losses due to criminals. A new AI-driven system has created realistic fake victims in the form of multilingual chatbots, wasting scammers' time and estimated to reduce losses by around $55 billion annually. Named after the Greek goddess of deceit, Apate, the system will use convincing cloned voices to engage in conversations with real scammers and "scam the scammers." Professor Dali Kaafar, Executive Director of the Macquarie University Cyber Security Hub, said, "Phone scams are operated by organized criminal groups, and currently only a small fraction of criminals are caught, with very little money recovered." Kaafar came up with the idea while having lunch with his family, when a scammer called. He pretended to be interested, making his children laugh and keeping the scammer on the phone for 40 minutes. "I realized that while I wasted the scammer's time, preventing them from reaching more vulnerable people, I also wasted 40 minutes of my own life," Kaafar said. "Then I started thinking about how to automate the whole process and develop a computerized chatbot using natural language processing to engage with scammers," he said. Kaafar said his team is currently applying for a patent for this technology. "We are excited about the potential of this new technology to actively disrupt the business model of scam phone calls and make them unprofitable," he said. Reducing scammers' profits According to Kaafar, phone scams are on the rise globally, with several factors contributing to this trend. Technologies like Voice over Internet Protocol (VoIP) allow cybercriminals to easily and cheaply disguise their location and pretend to call from any number. At the same time, updating telecommunications infrastructure and protocols to improve call authentication is difficult and expensive. "Economically, this is a high-profit, low-cost practice for scammers, which is very lucrative and a relatively low-risk criminal activity - it is difficult for victims to recover the money," Kaafar said. These factors attract more scammers who excel at exploiting human emotions and fears. "The scammers' business model relies on making huge profits from a small number of victims; among the thousands of calls they make each week, only a small fraction is successful," said Professor Kaafar. "We are constraining them, wasting their time, and reducing the number of successful scams. We can disrupt their business model and make it harder for them to make money." How APATE was created The team from Macquarie University Cyber Security Hub first analyzed scam phone calls and identified the social engineering techniques scammers use on victims, using machine learning and natural language processing to identify typical scam "scripts." They then trained the chatbot on real-world scam conversation datasets, including recordings of scam phone calls, transcripts of scam emails, and chat logs from social media platforms, enabling the chatbot to generate dialogues similar to real-world scam phone calls. Professor Kaafar said advancements in natural language processing (NLP) and AI human voice cloning technology enabled them to develop fluent-speaking AI agents that can adopt specific roles, stay on track during conversations, and provide consistent responses. "The conversational AI agents we developed can deceive scammers into thinking they are talking to viable scam victims, so they spend time trying to scam the AI agents," Kaafar said. These chatbots can be trained in any language or accent, as phone scams are a global challenge, and this technology can be deployed anywhere in the world. Live scam call trials The team is now testing the chatbot on real-time scam phone calls, redirecting calls intended for victims to their testing prototype, which acts as an "always-on honeypot" with various roles. "We spread these 'dirty' numbers on the internet, getting them into some junk applications or posting them on webpages to make them more likely to receive scam calls," Kaafar said. "We found that the chatbots perform well in some tricky situations, where we didn't expect them to escape, such as when scammers ask for information we haven't trained the chatbot on, but the chatbot is adapting and providing very credible responses." "The chatbots are continuously learning how to prolong the call to achieve their main goal: keeping scammers on the line for longer," he added. The average call duration for the deployed Apate chatbots has already reached 5 minutes, with the goal of reaching 40 minutes. Anti-scam chatbots also help provide threat intelligence - timely collection of information about current phone scams and their targets, which helps organizations such as major banks, retailers, and government agencies issue alerts to their customers. Professor Kaafar said the team is in discussions with some telecommunications providers and is open to some commercial partnerships. "Collaboration with communication providers will be key to making it truly effective," Kaafar said. "We believe this has enormous potential globally; if we can redirect the junk calls that providers currently intercept and send scammers to Apate chatbots, occupying their time as much as possible, it will have a significant impact on scammers." "I think the ultimate meta-scenario would be scammers themselves adopting AI and training their own scam chatbots, which are then transferred to chatbots owned by telecommunications providers," he said. "If scam chatbots end up talking to anti-scam chatbots instead of stealing money from real people, I would consider that a huge victory." Source: https://techxplore.com/news/2023-06-scamming-scammers-ai-created-fake-victims.html

RECENT AI NEWS

RECENT AI TOOLS