AI Is Getting Really Good at Messing With Cybercriminals
AI is increasingly being deployed to combat cybercrime by engaging scammers with bots, wasting their time, and collecting intelligence, offering a new approach to a persistent global problem.
Intelligence analysis by Gemini 2.5 Flash
Governments struggle to curb cross-border online crime, leading to innovative solutions like using AI. Companies are now deploying AI bots to interact with scammers, keeping them occupied and gathering crucial data, effectively turning the tables on cybercriminals.
Imagine bad guys trying to trick people on the phone to steal their money. Now, imagine clever computer robots pretending to be those people, talking to the bad guys for a long, long time, but never actually falling for the tricks. These robots are like super-smart actors who waste the bad guys' time, so they can't trick real people, and also learn how the bad guys operate.
Analysis
The global challenge of online crime has long outpaced traditional law enforcement efforts, primarily due to the borderless nature of digital scams. This has spurred a shift towards proactive disruption, with artificial intelligence emerging as a potent tool. The article details how AI is being engineered not just to identify threats, but to actively engage and frustrate cybercriminals, turning their own tactics against them.
Apate
An Australian company named Apate, after the Greek goddess of deception, has developed a sophisticated AI system designed to divert phone scammers. Their platform employs approximately 350,000 bots that answer scam calls, infiltrate online chat groups, and respond to text messages. These bots are trained to maintain conversations for extended periods, giving scammers false hope while never falling victim to their schemes. The primary goal is to squander the criminals' resources and time, thereby reducing the number of potential real victims they can target.
Apate's bots are equipped with diverse personalities, language skills, and profiles to enhance their convincingness, making it difficult for scammers to detect they are interacting with AI. This strategy has allowed the company to collect over 250,000 pieces of real-time intelligence, including scam URLs, money mule accounts, and bank details. This data is invaluable for banks and telecom companies that utilize Apate's services, providing actionable insights into the evolving landscape of cybercrime.
Kernel Panic
The effectiveness of Apate's system was put to the test by the Kernel Panic newsletter team, who engaged with a demo version of the AI. Playing the role of scammers, the testers found the AI 'victims' to be highly engaging and responsive, maintaining natural conversation timing. The bots were designed to express a healthy degree of skepticism while still providing enough openings to encourage scammers to persist, a dynamic that proved intensely frustrating for the human testers.
Despite diligent efforts over several minutes, the testers were unable to convince the Apate AI to 'invest' in a cryptocurrency opportunity. This hands-on experience underscored the bots' ability to convincingly simulate human interaction and resist scam attempts. Apate's CEO, Dali Kaafar, noted that calls with their bots regularly extend beyond two hours, demonstrating their success in tying up scammers' time and resources.
ETH Zurich
The application of AI in disrupting cybercriminals extends beyond direct engagement with scammers to include more subtle defensive strategies, such as enhanced honeypots. Mark Vero, a doctoral researcher at the department of computer science at ETH Zurich, highlights the increasing integration of Large Language Models (LLMs) into open-source honeypot providers. These LLM-powered honeypots are designed to appear more realistic and adaptive, making them more effective at luring and retaining malicious actors.
Research conducted by Vero and his colleagues demonstrated that LLM-simulated honeypots could keep AI agents attacking their systems significantly longer compared to honeypots with more predictable behaviors. The agentic attackers were more convinced by the LLM-powered systems and identified them as actual honeypots at a much lower rate. This suggests that well-built AI-driven defensive systems offer a considerable advantage to defenders by wasting attackers' time and gathering critical intelligence on their techniques.
Key points
- AI is being used to create bots that engage cybercriminals, wasting their time and resources.
- Apate, an Australian company, deploys 350,000 AI bots to answer scam calls and infiltrate chat groups.
- These bots collect real-time intelligence on scammers, including URLs and bank details, for banks and telecom companies.
- LLM-powered honeypots are also proving effective at deceiving and retaining cyberattackers for longer periods.
- The goal is to disrupt cybercriminal operations and protect potential victims by occupying scammers with AI interactions.
AI offers a promising new frontier in the fight against cybercrime, capable of significantly disrupting scam operations by wasting criminal resources and gathering vital intelligence. This proactive approach could lead to a substantial reduction in successful scams and foster better collaboration among law enforcement, banks, and tech companies.
Despite these advanced AI efforts, cybercriminals operate on an industrial scale, initiating billions of attacks annually, making it an uphill battle. The sheer volume and sophistication of these operations mean that AI tools, while effective, may only mitigate rather than eliminate the pervasive threat of digital scamming.
