20万AI当假客户耗骗子，还套情报反制诈骗

For years, anti-fraud efforts have stuck to the same old playbook: warn people not to be greedy, not to trust unknown contacts, not to click suspicious links. These warnings do offer some protection, but they can never outpace scammers’ core strategy—probability and scale. Scammers spend hundreds or thousands on bulk phone calls, knowing only a tiny fraction of recipients will fall for their tricks. Yet even one successful scam covers all their costs, and scammers can easily offset the impact of anti-fraud warnings by dialing more numbers, expanding their reach.

But recently, an Australian AI company has developed a far more effective approach to counter scammers, turning their own tactics against them.

Fraud operations, at their core, function like sales companies: they acquire customers via mass calls, convert leads through scripted pitches, and collect payments through dedicated accounts. The most scarce resource for these operations isn’t phone numbers or even scripts—it’s the time of the scammers themselves. A single fraudster can only dedicate genuine attention to a handful of potential victims in a day, as building trust and guiding someone through the scam process requires consistent effort.

Understanding this, the company behind the countermeasure decided to flood scammers’ “sales funnel” with fake customers—all powered by AI. These AI entities don’t just hang up immediately; they engage in full conversations, follow the scammers’ prompts, and even act like they’re on the verge of transferring money. The only difference? They never actually send any funds.

This tactic is designed to waste scammers’ most valuable asset: their time. Where a real scammer might move on after a few seconds of a hang-up, these AI entities keep the conversation going for minutes, even hours. A scammer who spends an hour talking to a fake client loses the chance to target a real person, and over time, this inefficiency starts to break their operations.

The company has built a large library of AI personas to make these interactions realistic. Initially, there were just 120 different characters, but by August this year, that number had grown to nearly 200,000. Each persona has unique traits: different ages, genders, accents, and personalities, complete with natural pauses, hesitations, and small verbal cues like “um” or “ah” to feel human.

The results have been striking. In October, industry reports noted that a local telecom provider had routed nearly 590,000 suspicious calls to the company’s AI system. The AI entities engaged in over 2.5 million conversations with scammers, a total of time equivalent to one person staying awake and talking nonstop for 17 years. Many scammers grew frustrated and even angry, with the company eventually adding “number of times scammers use abusive language” as a performance metric—more abuse meant the tactic was working.

But the company’s strategy doesn’t stop at wasting time. It also uses these AI interactions to gather intelligence on scammers’ operations, turning the tables to collect critical data that helps stop fraud at its source.

To do this, the AI personas pose as potential victims in online spaces where scammers operate—investment forums, cryptocurrency groups, and social media platforms. When a scammer initiates contact, the AI follows their lead, asking questions that guide the conversation toward payment details: bank account numbers, cryptocurrency wallet addresses, and information about the “money mule” networks scammers use to move funds.

Scammers rely heavily on these mule accounts, as they need to constantly replace real accounts that get flagged by banks. Setting up these accounts is expensive and time-consuming, so scammers are eager to use any valid account offered. In one case, the AI found an online market recruiting Indian bank accounts for fraud, with brokers offering 5% of the fraud proceeds as commission, paid in cryptocurrency.

In another instance, an AI entity spent four days talking to a single scammer, collecting 10 different Australian bank accounts linked to the same fraud network as the scammer kept looking for new accounts. This data gives banks a critical edge: instead of waiting for victims to report fraud after their money is stolen, banks can use this intelligence to block transactions before they go through.

Major banks have already recognized the value of this approach. One of Australia’s largest commercial banks has been working with the company since 2024, expanding their partnership in 2025. The bank reports that the intelligence provided by the AI system is integrated into their anti-fraud tools in near real time, helping identify new scams and block transactions before they succeed.

For banks, this is a cost-effective solution. One major Australian bank spends over 1 billion Australian dollars annually on fraud prevention, nearly 10% of its annual profit. Each intercepted fraud saves the bank from payouts, investigation costs, and reputational damage. The AI intelligence also has a much lower marginal cost than traditional manual investigation methods.

This creates a closed loop: AI wastes scammers’ time, scammers accidentally reveal their operations’ details to the AI, banks pay for this intelligence, and the cycle continues. By August, the company had collected over 250,000 pieces of fraud intelligence, all used to support anti-fraud efforts.

Capital has also taken notice: in August, the company raised 8.15 million dollars in seed funding, and has since expanded its team and plans to expand to other regions like London and North America.

Perhaps the most ironic part of this strategy is that the intelligence used to fight fraud comes directly from the scammers themselves. Scammers spend their days perfecting pitches and targeting victims, only to have their own tactics, accounts, and networks exposed by a simple AI. They reveal all their secrets while thinking they’re close to stealing money, leaving them empty-handed in the end. For scammers, the only hope left is that the next person they talk to is a real human, not another AI.