AI can help scammers reach potential victims at scale, but it can also be used to slow them down. The Australian company Apate has built bots that answer calls, messages and chats, keeping fraudsters engaged while collecting information about their methods.
Keeping fraudsters on the line
Apate has spent the last two years developing a system that diverts phone scammers to AI personas. The bots are designed to sustain a conversation without falling for the pitch. The longer a scammer stays occupied, the less time they have to contact other people.
Apate founder and CEO Dali Kaafar says the company wants to create “the perfect victims for scammers.” The platform is used by banks and supported by telecom companies. Kaafar says it has around 350,000 bots, which also join scam chat groups and respond to text messages.
The conversations serve a second purpose: gathering intelligence. Kaafar says Apate has collected more than 250,000 pieces of information about fraudsters in real-time, including scam URLs, money mule accounts and bank details. Such details can help build a picture of how scams operate and where they may lead.
Personas built to seem real
The bots use different personalities, language skills and profiles, according to Kaafar. Some may have WhatsApp; others may not. They may answer a call or end it with a promise to return later. Those variations are meant to make the interactions less predictable to scammers.
A demo tested by Kernel Panic let users play scammers trying to persuade an AI “victim.” The persona showed some skepticism while leaving room for the conversation to continue. The system responded naturally enough to keep the test going, but it did not agree to invest in a cryptocurrency opportunity during six minutes of effort.
Kaafar says Apate calls can last more than two hours. Even a single prolonged interaction could tie up a scammer who might otherwise use automated dialing tools to reach many people. The approach depends on sustaining the fraudster’s interest without exposing that the apparent target is an AI.
AI expands the honeypot approach
Apate is part of a wider effort to disrupt cybercrime by using automation against criminals. Governments have struggled to address online crime across borders, while law enforcement and research efforts have not stopped digital scamming from expanding. The source notes that some sophisticated operations use physical, industrial scale scamming sites.
Security teams have long used honeypots: false virtual machines designed to attract hackers, waste their time and reveal techniques. Mark Vero, a doctoral researcher at the department of computer science at ETH Zurich, says open source honeypot providers are increasingly adding large language models (LLMs) to make these systems seem more realistic.
In research, Vero and colleagues found an LLM-powered honeypot kept AI agents attacking significantly longer than a honeypot with more predictable behavior. The agents were also less likely to identify the simulated systems as honeypots. More convincing decoys may therefore give defenders extra time and information, provided the systems are built well enough.
Disruption has limits
AI-driven decoys do not mean scam calls and texts will disappear. Cybercriminals initiate billions of messages and calls each year, and the source describes the overall expansion of digital scamming as continuing despite deterrence work by professional scambaiters and initiatives to infiltrate scam operations.
Still, keeping criminals busy and learning from their attempts could strengthen defenses. Better intelligence sharing among police, social media companies, banks and other industries is needed, and AI monitoring and data analysis could help. The goal is not only to detect scams, but to make criminal operations spend time and resources while their methods become clearer.