AI Honeypots Turn the Tables on Cybercriminals

Overview

As artificial intelligence increasingly empowers cybercriminals to execute sophisticated scams at scale, a counter-movement is emerging that weaponizes the same technology against attackers. Governments and security firms are turning to automated systems designed not to prevent crime, but to entrap, delay, and gather intelligence on fraudsters.

This shift marks a departure from traditional defensive strategies, which often struggle to prosecute offenders across international borders. Instead, innovators are deploying AI-driven honeypots that act as the perfect victims—keeping scammers engaged while siphoning valuable data about their operations.

The Apate Strategy: Building Perfect Victims

In Australia, a company named Apate has spent the last two years developing a system that diverts phone scammers onto calls with AI bots trained to sustain conversations indefinitely. The goal is simple yet effective: waste the scammer’s time and resources while collecting real-time intelligence.

“We’re building the perfect victims for scammers,” says Dali Kaafar, founder and CEO of Apate. “A minute that a scammer is talking to a bot is a minute where you’re probably saving hundreds, if not thousands, of possible people.”

The platform, which leverages approximately 350,000 bots, is currently utilized by banks and supported by telecom providers. These agents do not merely answer incoming calls; they also infiltrate online scam chat groups and respond to text messages. Their primary directive is to frustrate criminals by appearing interested without ever falling for the ruse.

To ensure they do not blend in with automated scripts, these bots exhibit varied personalities and language skills. Some may pick up the phone, while others might hang up prematurely or claim they will return the call later, mimicking the unpredictable nature of human behavior.

From Theory to Frustration

The efficacy of such systems was tested during a demonstration where participants assumed the role of scammers attempting to defraud an AI “victim.” The personas were designed to express skepticism while leaving enough openings for the attackers to persist.

For the testers, the experience was intensely frustrating. Despite tag-teaming as both a friend and a financial advisor, they could not persuade the AI to invest in a fake cryptocurrency opportunity within six minutes. The system remained engaged, responsive, and strategically vague.

Kaafar notes that real-world calls frequently extend beyond two hours. During this time, the platform has collected more than 250,000 pieces of intelligence, including scam URLs, money mule accounts, and bank details. This data is provided in real-time to authorities and financial institutions, potentially disrupting entire fraud networks.

However, the scale of the problem remains daunting. Cybercriminals initiate billions of messages and calls annually, with sophisticated operations running physical sites of industrial scale. While initiatives like Apate provide critical deterrence, they represent a fraction of the total volume of malicious activity.

Expanding the Battlefield: LLM-Powered Honeypots

The use of AI in defensive cybersecurity is not limited to consumer fraud. Security researchers have long employed honeypots—false virtual systems designed to attract hackers, waste their time, and analyze their techniques. Now, these systems are becoming significantly more realistic through the integration of Large Language Models (LLMs).

Mark Vero, a doctoral researcher at ETH Zurich, recently collaborated on a study examining how LLM-powered honeypots perform against AI-driven attackers. The findings suggest that these advanced systems can keep aggressive agents engaged for significantly longer periods than traditional, predictable honeypots.

“The agentic attackers are much more convinced by the LLM simulated honeypots and they also mark them as actual honeypots at a much lower rate,” Vero explains. “If these systems are built well enough, then I think it’s quite advantageous for defenders.”

This evolution in defensive tactics highlights a broader trend in cyber warfare: exploiting the psychological vulnerabilities of attackers. Beyond technical deception, some efforts now include direct psychological operations, such as trolling ransomware operators to disrupt their workflow and morale.

A New Era of Digital Defense

The rise of generative AI has created a symmetrical battlefield where both sides utilize sophisticated automation. For defenders, the strategy has shifted from pure prevention to active disruption. By occupying scammers’ time, anti-fraud AI reduces the bandwidth available for launching new attacks against other potential victims.

Success in this domain will likely depend on improved intelligence sharing between law enforcement, social media companies, and financial institutions. AI-driven monitoring can bridge these gaps, providing a unified view of criminal networks that operate across multiple platforms and jurisdictions.

While no single tool can eliminate digital crime, AI honeypots represent a promising avenue for turning the tables. As cybercriminals grow more dependent on volume and speed, any mechanism that slows them down and reveals their identities becomes a critical asset in the ongoing fight for digital security.