An Australian Startup Sent 350,000 AI 'Victims' Into Scam Compounds' Phone Banks. Chinese Readers Made It a Hero.
Apate, a Sydney anti-fraud startup, trains AI agents to waste scammers' time by posing as ideal marks — 9 million automated conversations and counting. The idea has gone viral in China, a country with its own painful scam-compound epidemic, and inbounds are flooding in from Chinese banks and platforms.
The nickname Chinese social media chose for it is instantly legible: “缅北最严厉的父亲” — the strictest father of northern Myanmar, the region whose industrial-scale scam compounds have swallowed thousands of Chinese jobseekers. The object of the affection is Apate, a Sydney startup spun out of Macquarie University's cybersecurity lab, whose founders did something no blocklist ever thought of: instead of stopping scammers from reaching victims, they built 350,000 AI agents designed to be the victims.
The company's co-founder Peter Eckermann told China News Weekly this week that in the past week inquiries have poured in from China — big social platforms, telecom carriers, commercial banks. “I have to pinch myself to believe all this is real,” he said. The mechanics are almost mischievous. The agents answer scam calls, then follow the script the fraudster sets: move to WhatsApp, open the email, click the phishing form, keep chatting. They also seed their fake profiles into suspect web forms and dark-web leaks, so the marks come to them. Per the company's figures, its agents have held more than 9 million fully automated conversations, burning a cumulative 17 years of scammers' time; the average call lasts 15 to 20 minutes, and even a two-minute hang-up is a tax on the caller.
Does it actually work? Only about 2 percent of scammers realise they are talking to a machine, Eckermann claims — partly because the demo voices on the website are deliberately old versions, and partly because the decisive element isn't the voice at all. The system is built on game theory: each conversation is modelled in stages (when the lure is dangled, when the payment is demanded) and reinforcement learning tunes when agents reward or stall. The firm, named after the Greek goddess of deceit, has raised close to 10 million US dollars with about 20 employees.
The reason the story caught fire in China is not that the technology is exotic, but that the target is personal. For two years Chinese reporting — including undercover interviews with women trapped in scam compounds — has kept the human cost of Southeast Asian fraud industrial parks in front of readers, and Chinese police have run their own campaigns against telecom fraud. A machine that turns the scammer's phone bank into the victim's plaything flipped the script in a way readers found satisfying: the trend climbed the Baidu hot board under the strictest-father nickname within a day. “For the last thirty years we've been fighting fraud in completely the wrong way,” founder Dali Kaafar told the magazine. On Chinese social media this week, few were inclined to argue.