For years, the world has been locked in a battle against fraud, yet the tactics employed by scammers remain largely unchanged.

We're told to curb our greed, exercise caution, and avoid clicking on suspicious links.

Do these precautions work? To some extent, yes.

However, scammers don't rely solely on persuasion; they bet on probability. They make countless calls daily, with the vast majority going unanswered. Yet, if just one person responds, believes their lies, and transfers money, their daily expenses are covered.

You might successfully dissuade a hundred people with your earnest advice, but scammers can compensate by making a thousand more calls. This is why anti-fraud campaigns often struggle to keep pace with scammers.

Recently, however, an Australian AI company devised a clever strategy.

Apate founder Dali Kaafar (front and center) with team members

Since scammers profit from volume and probability, why not inundate their pool of potential victims with fake clients, all portrayed by AI?

These AI clients can answer calls, engage in conversation, follow scammers' instructions meticulously, and even feign the act of transferring money.

The only difference from real people? They never actually pay.

Scammers believe they've landed a big catch, but after hours of conversation, they realize they've wasted their time.

What's even more impressive is that these AI clients can also extract crucial information from scammers, such as bank account details and cryptocurrency wallet addresses, which they then provide to banks for anti-fraud purposes.

Today, let's delve into this remarkable company.

/ 01 / 200,000 AI clients engage scammers until they're exhausted

The fraud industry may seem enigmatic, but at its core, it operates much like a sales company.

Customer acquisition relies on mass calls, conversions hinge on scripts, delivery depends on payment accounts, and revenue is generated through profit-sharing.

In this model, the most costly element is human labor.

Phone numbers can be purchased in bulk, scripts can be replicated, and now, even voices can be cloned.

However, the process of transitioning a person from "answering the call" to "transferring money" involves building trust, explaining procedures, and soothing emotions—tasks that machines cannot yet replicate.

Thus, the most scarce resource for scammers is their employees' time. A scammer can only handle a few clients seriously per day.

Once you grasp this, Apate's strategy becomes clear.

It employs an intriguing tactic: flood the scammers' sales funnel with fake clients to dilute their efficiency.

Faking traffic is a tactic familiar to internet companies. In the past, if monthly reports were lackluster, they would purchase traffic to inflate unique visitor (UV) counts. Apate is doing the same, but its adversary is now a scamming syndicate.

It partners with Australian telecom operator TPG, which identifies suspicious calls and redirects them directly to Apate's AI.

Numbers that ordinary people would hang up on, block, or mark as spam are treated as treasures by Apate. Not only that, but it also releases a plethora of virtual numbers online, waiting for scammers to take the bait.

When scammers call, an AI always answers. And these AI clients are highly cooperative, representing high-quality fake traffic.

Most importantly, they don't just hang up; they keep the conversation going.

Hanging up would be too easy for scammers. It would only waste a few seconds, and scammers would simply curse and dial the next number.

Apate aims to keep scammers on the line indefinitely. For scammers, the most precious commodity isn't the phone bill; it's the time they could have spent scamming real people during those tens of minutes.

Apate's official website released a genuine recording. A scammer poses as technical support from Australia's National Broadband Network (NBN), claiming there's an issue with the user's home network that requires remote repair.

On the other end of the line is a very cooperative lady.

When asked to check the router, she reports the indicator status. When asked to test the internet speed, she follows the instructions. Then, the scammer reveals his true intention: to download remote control software AnyDesk.

Once authorized, the other party can directly operate your computer, accessing online banking, passwords, and initiating transfers.

The critical moment arrives. The lady starts to falter. She claims she can't find the browser, worries the software will incur a fee, and complains about having to repay her car loan next week and being short on cash.

Finally, she starts downloading but says the progress is too slow and goes to pour water.

The recording ends without the scammer connecting.

This is Apate's ideal client for scammers: appears ready to pay but never actually does.

To make this scenario realistic, Apate constructed a virtual character library.

It began with only 120 personas. By August of this year, it had expanded to 197,000, nearing 200,000.

Each persona has a distinct age, gender, accent, and personality. Even the pauses, hesitations, and unconscious sounds like "um" and "ah" during thought are meticulously designed.

200,000 fake clients, available 24/7, never transferring money.

The impact was immediate. In October of this year, TechCrunch reported that TPG had redirected nearly 590,000 scam calls to Apate.

Apate revealed that its AI had engaged in 2.5 million conversations with scammers.

These hours accumulate to the equivalent of one person chatting non-stop with AI for 17 years without eating or sleeping.

It's quite surreal.

Scamming syndicates invest in numbers, employees, and script training, striving to improve conversion rates, only to waste a significant amount of time on non-existent clients.

Some scammers lose their temper after chatting for a while and start cursing.

Later, Apate even incorporated the "number of times scammers curse" into its key performance indicators (KPIs). The more they curse, the worse the scammers are being tormented.

/ 02 / Scammers: AI's Unwitting Intelligence Agents

But if the story ended here, Apate would merely be an advanced version of a nuisance call.

Apate's even more potent move is its ability to extract intelligence from scammers.

Specifically, it has AI impersonate potential investors with virtual identities, infiltrate investment forums, cryptocurrency groups, and social platforms, and actively approach scammers.

Once scammers take the bait, the AI continues the conversation following their scripts, from investment returns to transfer procedures, gradually extracting key information like the bank accounts and cryptocurrency wallet addresses they intend to use.

This tactic succeeds because there's an inevitable link in the fraud supply chain: receiving money.

Scammers can change numbers, scripts, and identities at will. They can also switch receipt accounts, but at a much higher cost.

After all, bank accounts require opening, identity verification, and someone to receive and transfer funds. Once an account is flagged by risk control systems or even frozen, scammers must find new receipt channels.

This has given rise to a specialized underground business: money mule accounts.

Simply put, someone provides their bank account to scamming groups to receive and transfer money, earning a commission in return.

In July of this year, Apate's AI uncovered an online marketplace specifically recruiting Indian-verified bank accounts. Intermediaries were even willing to offer 5% of the fraudulent funds as a commission and settle in USDT.

More intriguingly, Apate disclosed another case.

The AI had been engaging with a scammer for four whole days. During this time, the scammer kept seeking new receipt accounts, and the AI recorded each one. Ultimately, the AI obtained 10 Australian bank accounts associated with the same scamming network from a single conversation.

This exposes another facet of the fraud supply chain. While receipt accounts can be changed, a continuous supply network for recruiting, providing, and managing accounts is needed behind the scenes.

And Apate is exploiting scammers' eagerness to receive money to extract this account information in advance and provide it to banks for risk identification and intervention.

The value of this to banks is self-evident.

Consider this. In the past, banks often had to wait for victims to realize they'd been scammed and report it before they could track the money flow. But by then, the money was long gone.

Now, if they know the card numbers and wallet addresses before scammers receive the money, they have a chance to intercept the transactions midway.

So, who foots the bill for this intelligence? The answer is banks.

The Commonwealth Bank of Australia (CBA), the country's largest commercial bank, is Apate's client.

They initiated a pilot in 2024 and announced an expanded partnership in June 2025. According to the bank, Apate's intelligence can be integrated into the anti-fraud system almost in real-time, aiding in identifying new scams, generating warnings, and intercepting fraud.

Why are banks willing to pay? Because the math is straightforward.

CBA alone spent over 1 billion Australian dollars on preventing fraud, scams, financial crimes, and cybersecurity in the 2026 fiscal year. That's nearly 10% of its annual net profit.

For banks, anti-fraud has long been an ongoing operational cost. Every additional fraud intercepted reduces customer losses and subsequent compensation, investigation, and disposal costs.

Moreover, the marginal cost of AI obtaining this intelligence is far lower than traditional methods relying on manual investigations.

Thus, the loop is closed. AI tricks scammers, scammers voluntarily provide intelligence, and banks and telecom operators pay for it.

Public data from August of this year shows that Apate has extracted over 250,000 pieces of fraud intelligence, all entering its clients' disposal processes.

Capital has also caught wind of this. In August of this year, Apate completed an 8.15 million USD seed funding round led by Lobby Capital.

In October reports, the company had only about 20 people on its team. Its registered entity had already relocated to the US, and it planned to open an office in London, preparing to expand across Europe and America.

The most punk aspect of this entire business is that its intelligence source is precisely the scammers themselves.

Scammers clock in every day, memorize scripts, chase quotas, and try their utmost to scam money from others.

But after hours of conversation, they not only fail to secure any money but also reveal their receipt channels, bank accounts, and even the profit-sharing rules of the fraud supply chain to a robot, sentence by sentence.

Ironically, scammers voluntarily share this information.

Because in their eyes, they're conversing with a client who's about to transfer money. With just a few more words and a little more patience, the money will be theirs.

In the past, scammers profited by exploiting others' trust. Now, Apate exploits scammers' greed to extract intelligence from them and sells it to banks.

In this entire business chain, scammers contribute time, scripts, and intelligence but end up with nothing.

Now, the only thing these scammers can do is pray every day when they start work:

"At least the client who's been hesitating to transfer money today is a real person."

By Yuanyuan