Fighting Fire With Automation
For years, governments have struggled to chase online criminals across borders, where jurisdictions are fragmented and enforcement moves slowly. That gap has pushed defenders toward a different idea: rather than only catching scammers, make their work as costly and frustrating as possible. Now generative AI is turning that idea into working infrastructure. The result is an unusual arms race in which machines are being deployed to waste the time of other machines and the humans behind them.
Consider the Australian company Apate, which takes its name from the Greek goddess of deception. Its system routes phone scammers to AI personas designed to sound skeptical but never fully convinced. According to its founder, the platform runs around 350,000 bots and calls have been known to stretch past two hours. Every minute a scammer spends talking to a bot is a minute they cannot spend dialing a real person. That arithmetic matters when criminal operations depend on automated dialing tools that can reach thousands of potential victims in a single session.
Why Believable Victims Matter
The central challenge is realism. A bot that answers too quickly, repeats itself, or falls for an obvious ploy is useless because scammers will simply hang up and move on. What makes these systems effective is the variety of personalities, language skills, and behaviors layered on top of large language models. Some personas have WhatsApp, some do not. Some hang up with a promise to call back. That unpredictability is what keeps a scammer invested. Early testing suggests the approach works. Even determined users trying to pitch a fake cryptocurrency opportunity to a bot persona found the AI resistant, which is precisely the point.
The same logic applies beyond phone calls. Research from ETH Zurich has found that honeypots built with large language models keep AI-driven attackers engaged significantly longer than static decoys, and attackers are less likely to identify them as traps. As criminals themselves adopt agentic AI to scale their operations, defenders who can match that sophistication gain a real edge. The question is no longer whether machines will fight machines, but which side builds the more convincing deception.
What Comes Next for Defenders and Consumers
Deception-based defense is not a complete answer. Billions of scam messages still go out each year, and no bot farm can absorb them all. Better intelligence sharing between banks, telecoms, social platforms, and police will be essential, and AI-driven analysis could help connect those dots faster than human teams can. Yet the trajectory is clear: generative AI is becoming as useful for protection as it is for attack, and the organizations that adopt these tools early will set the standard. For everyday people, this shift should eventually mean fewer scam calls reaching them, and stronger protections embedded in the banking and telecom services they already pay for. Consumers evaluating smartphones, security subscriptions, or fraud-monitoring services should watch for vendors that openly describe how they use AI in threat defense, because that capability is quickly moving from novelty to a basic expectation in the products people choose to buy.
