Australian technology firm Apate is deploying a vast array of AI-bot characters around the world that play the role of gullible scam victims, wasting millions of hours of con artists' time every month.

Among the company's monthly performance metrics is an unusual figure: how many times frustrated scammers swear at the oblivious "victims" who play dumb and string them along.

"I think we're the only company in the world that is actually keeping as part of their KPIs the number of F-words that scammers are dropping at them," Apate founder Dali Kaafar tells Magazine with a chuckle.

The company maintains a stable of almost 200,000 AI characters capable of holding convincing phone conversations and chatting on social media and messaging platforms. "I can tell you that we're basically servicing, as we call them, hundreds of thousands of calls a day, and pretty much hundreds of thousands of conversations on the other channels," he says.

Every hour of a con artist's time that the bots waste is another hour the fraudster is not scamming a member of the public. In the six weeks up to the end of 2025, Apate's bots engaged in 600,000 scam calls for a single Australian telco called TPG.

"Essentially, we wasted more than five hundred days of scammers' time," he explains. "That roughly equates to somewhere around thirteen million dollars being saved."

That is meaningful money in a country where scams are a multi-billion-dollar problem: Australians reported more than $2 billion in losses to the consumer watchdog's Scamwatch service in 2024 alone, a figure regulators caution understates the true toll because many victims never file a report.

The bots' other goal is to elicit actionable intelligence for banks and telcos combating scam rings across Australia, Asia, Africa, the UK and Europe.

Scam baiting at scale with AI victims

Kaafar says the idea came to him when he received a scam call while on a picnic with his family in Sydney in November 2021. To his wife's annoyance, but his kids' delight, he strung the scammer along for 44 minutes by pretending to be a gullible rube falling for the con.

"What followed was really literally a full comedy show for my kids," he says, adding that during the call he also learned a great deal of potentially useful information about the mechanics of the scam and the tactics used.

"As I hung up that call, I remember thinking very clearly: if I could do that just for fun, imagine what technology can do at scale."

A professor at Macquarie University at the time, he raised the idea with some of his doctoral students working on AI and security.

"I said, 'Guys, there has to be a much better way of doing this. Let's build something that is really automating this whole process of engaging scammers at scale, but also, perhaps most importantly, extracting all sorts of intelligence from these conversations.' And that's literally how it started."

Within a few months the team secured funding from the Office of National Intelligence to research the idea, and the project was spun off from the university into Apate in 2023, taking its name from the ancient Greek goddess of deceit. The company now works with most of the big banks in Australia, as well as numerous banks in the UK, South Africa and South East Asia.

Demand for such tools is being reinforced by regulators. The UK's Payment Systems Regulator has required banks to reimburse victims of authorised push payment scams — up to £85,000 per claim — since October 2024, and Australia's parliament passed a Scams Prevention Framework in 2025 that phases in binding anti-scam obligations for banks, telcos and digital platforms.

Apate is far from the only company scam baiting fraudsters using AI bots, though it operates on a greater scale than most. United Kingdom telco O2 rolled out an AI Granny campaign last year — a chatty pensioner named Daisy, built with input from veteran YouTube scambaiter Jim Browning — which frustrates scammers by taking up hours of their time talking about her 28 cats. The initiative was as much an advertising campaign to warn the public about the dangers of scam calls as anything else.

Creating the perfect AI victims

Apate launched with 120 different personas across different genders, ages and personality types, and now has 197,000 personas with identifiable vocal tics and accents — and they even make the same noises people make when trying to think of what to say.

"We spent a lot of time refining and building these AI bots that sound exactly like you and I and our neighbors," Kaafar says.

The AI models were trained on "hundreds and hundreds" of hours of recorded conversations between human scam baiters and scammers, allowing them to employ counter strategies.

"They know that they're talking to bad guys, if you like, and they really navigate the conversations so that it sounds really very, very realistic to any scammer out there, even if a scammer is skeptical about things."

The bots are sent out on WhatsApp and Telegram, where they act as honeypots for scammers. While the old cliché that you can't scam an honest man is not true, it remains very possible to exploit a scammer's desire for money.

"They are cybercriminals, really. I think we just very often forget that they're cybercriminals who are trying to get people's life savings. And so that element of greed is sometimes what our bots also exploit."

Bots collect valuable intel from each scam call

In the crypto industry, Apate works for "one of the leaders in blockchain analysis," which may or may not be Chainalysis. That client is not primarily interested in wasting scammers' time — it wants intelligence on which wallets and methods scammers are using so they can track the flow of funds.

"These bots, as they engage across different conversations, extract new crypto wallet addresses by the hundreds and by the thousands," he says. "It's data and intelligence that is coming literally before their damage happens."

Scamming is big business, and the call centers behind it are "pretty much corporate organizations," Kaafar explains.

"This data is very, very important because, literally, that's the new account or the new wallet where you really need to pay extreme attention to. Because this is where these… scammer compounds are collecting their money or their funds with."

Such compounds are industrial-scale fraud operations, many of them based in South-East Asia, where the UN Office on Drugs and Crime estimated in 2023 that at least 220,000 people had been trafficked and coerced into running online scams — a measure of the organized machinery behind the calls Apate's bots field every day.

"Think about it literally as being always ahead of the scammer's tactics. And the more you know before the money gets transferred, the better it is."

In July, Apate's bots uncovered a marketplace for brokers soliciting verified bank accounts in India, offering commissions of up to 5% — paid in USDT — on the proceeds from scams passing through the accounts.

AI arms race between good guys and bad guys

Scammers are increasingly using AI bots themselves, and it will not be too long before AI scammers are as ubiquitous as spam emails. Scamming people is a $1.24 trillion business, so the industry can afford the compute required to scale up operations. In the United States alone, the FBI's Internet Crime Complaint Center recorded $16.6 billion in reported fraud losses in 2024, a rise of roughly a third on the previous year.

Apate's research suggests that about 20% to 30% of scam text conversations already employ AI, but Kaafar is not too worried about the outcome of anti-scam bots fighting scam bots. He says Apate's researchers believe AI bots playing defense have an advantage, according to game theory, because they are trying to extract intelligence, while the scam bots are trying to get the other AI to perform an action.

"You can also demonstrate mathematically that that is to the advantage of a defender because it becomes easier to extract intelligence from the attacker's AI model," he says.

"We can imagine a world where scammers become a lot more sophisticated and deploy such technology. But that also means that if they do, they're actually deemed to lose the game, which is great news in the fight against scams."

Related: AI-powered bot stands in for absent candidate in Virginia debate | Agent wastes 14 hours of scammers' time, LLMs 'poisoned' by Iran — AI Eye