AI agents are sent to be scammed so they can learn how fraudsters operate

Australian startup Apate.ai is deploying thousands of AI agents that pose as unsuspecting victims — answering scam calls, replying to suspicious messages, clicking phishing links and joining social-media groups — to lure fraudsters into extended conversations and harvest intelligence. The company, founded in 2024 on research from Macquarie University in Sydney, says its fleet has so far wasted 17 human-years of scammer time across 2.5 million interactions.

The agents adopt diverse personas — different ages, genders, accents and languages — and try to keep a scammer on the line long enough to reveal details that typically surface only after a real victim has been fleeced: bank accounts, crypto wallets, phone numbers, malicious websites, the brands the scammer is impersonating, and other specifics of the particular scheme. Apate feeds that data to its customers — banks, telcos, government agencies and infrastructure operators — which integrate it into their prevention systems.

The model arrives as AI-enabled fraud surges. The FBI recorded more than 22 thousand complaints tied to AI-related crime in 2025, with cumulative losses of roughly $893 million. About $632 million of that came from investment scams in which criminals use language models to run conversations and generate deepfake audio and video of celebrities, executives and other trusted figures.

The idea began when Apate founder and chief executive Dali Kaafar took a scam call during a family picnic. He decided to "play the innocent" for his children's benefit and kept the fraudster talking for 44 minutes. Kaafar realized those minutes were time the scammer wasn't spending on a real victim, and asked whether the tactic could be automated while simultaneously learning something useful. The machine-learning and natural-language-processing research he led at Macquarie, focused on detecting scam-call content, eventually became the commercial product.

Co-founder and chief commercial officer Brad Joffe said Apate's agents now handle tens of thousands of conversations a day with scammers across phone, messaging, email and web channels. The company blends state-of-the-art off-the-shelf models with internally developed models fine-tuned on the agents' own conversations and on Kaafar's academic research. Customers gain access to the agent fleet and a dashboard that surfaces the insights extracted from those interactions.

Apate reached the finals of the Startup Battlefield 200 competition at TechCrunch Disrupt this week in San Francisco, where it will present to a panel of investors.