Most companies build AI agents to get work done faster. Apate.ai builds them to get scammed, and then charges banks for the paperwork. The Australian startup runs thousands of agents that pose as ordinary victims, and its pitch is that the best time to learn about a scam is before anyone loses money.
A very odd job description
Apate was founded in 2024, on research done at Macquarie University in Sydney. Its agents answer scam calls, reply to suspicious messages and emails, click phishing links, and join social media groups to attract fraudsters. They come in many varieties, with different ages, genders and accents, and they speak a range of languages.
The goal is not to win an argument with a criminal. It is to be a convincing, slightly gullible person for as long as possible, because a scammer who keeps talking tends to keep revealing things.
The company says its agents have so far wasted 17 years of scammers' time across 2.5 million conversations. That is a satisfying number, but it is not the product. The product is what falls out of those conversations:
- Payment details: bank accounts and crypto wallets.
- Contact points: phone numbers and malicious websites.
- Context: the brands being impersonated and other particulars of the scam in progress.
Normally this kind of information surfaces only after a victim has already been robbed. Apate's customers, which include banks, telecom companies, government agencies and utilities, get it earlier and use it to shape their anti-fraud work.
Why now
The timing is hard to call accidental. Scammers have good tools too. The FBI received more than 22,000 complaints about AI-related crimes in 2025, with losses of about $893 million. Of that, more than $632 million came from investment scams, where, according to the FBI, fraudsters use AI to run conversations and to fake the voices and faces of celebrities, executives and other trusted figures.
Scam-baiting has long been a hobby for people who enjoy stringing fraudsters along. Some have got scammers to read entire books aloud, or even to get ridiculous tattoos. Apate is the industrial version, with the volunteer replaced by a fleet. The tables, as the saying goes, are being turned, though in this case with a sales team attached.
The 44-minute picnic
The idea started with a real scam call. Several years ago, Apate's founder and CEO Dali Kaafar was at a family picnic when an unknown number rang. He realised it was a scammer, decided to play dumb for the amusement of his kids, and kept the call going for 44 minutes.
Everyone laughed. Kaafar also noticed something less funny and more useful: for those 44 minutes, the scammer was not talking to a real victim. That raised an obvious question. Could he automate the stunt, and learn something along the way?
The answer began as research at Macquarie, where Kaafar and colleagues used machine learning and natural language processing to uncover what scam calls actually contain, and to find better ways to detect and prevent scams.
How the business works
Co-founder and chief commercial officer Brad Joffe describes the gap Apate wants to fill. Most of the industry, he says, works after the fact. Telcos block numbers, but scammers get new ones. Banks watch transactions, but by then the scammer has already reached the customer. Apate goes to the scammer directly, he says, and gets the information before anyone is hurt. He claims the agents talk to tens of thousands of scammers every day across calls, messages, email and the web.
Under the hood, Joffe says the agents and the analysis run on a mix of frontier models and Apate's own fine-tuned models. The company trains those models on the conversations its agents have had with scammers, along with Kaafar's research from Macquarie.
Customers get two things:
- The agents themselves. Customers can assign missions, such as obtaining a particular bank account number or crypto-wallet address. The text agents can also search forums, investment groups and social platforms for scam activity, then enter conversations under a fabricated identity to draw out the scammer's details.
- An insights portal. It offers intelligence and analytics on scams aimed at a customer's end users, employees, brands, products and more.
The customer results, as the company tells it
Apate's named customers include Commonwealth Bank of Australia and TPG Telecom, plus other utilities, telcos and government agencies. Joffe supplied three results. All are company figures, and none came with independent verification in the report.
| Customer | Reported result |
|---|---|
| TPG Telecom | Nearly 590,000 calls diverted to Apate's agents |
| Commonwealth Bank of Australia | 76% drop in publicly reported scam losses |
| Unnamed Australian bank | 22 scam campaigns disrupted using outbound agents |
The Commonwealth figure is the boldest, and it is worth reading carefully. It describes publicly reported scam losses, which is a narrower thing than all scam losses. The source does not say how much of the drop is down to Apate versus everything else a large bank does about fraud.
Money and plans
Investors are betting that fooling scammers pays. Apate raised $8.15 million in seed funding in August from Lobby Capital, OIF Ventures, Investible, Concept Ventures and Baobab Ventures. Joffe says the 20-person company has raised just under $10 million in total. The new money is earmarked for product work and international expansion, including a London office and a larger presence in the US.
Apate was also a finalist in the Startup Battlefield 200 at Disrupt in San Francisco, where it was set to pitch a panel of VC judges. The event ran at Moscone West through Thursday, October 15.
A goddess, and some swearing
The name comes from Apate, the Greek goddess of deception, which Joffe says felt fitting given what the agents do. He adds that the pitch material even includes recordings of scammers swearing at the bots.
There is a method to it. The company tracks swearing as a sign of frustration, and celebrates every instance. Joffe reckons Apate is probably one of the only businesses in the world that cheers when people curse at its product.
It is a rare customer-satisfaction metric in which the angriest users are, by design, the criminals.