📰 Key Takeaways

As scammers increasingly run into AI-powered defense tech, Australian startup Apate trains AI bots to pose as victims and fight back against scam operations. Founder Professor Kaafar says the company’s bots handled 600,000 scam calls for a single telco, TPG, in just six weeks before the end of 2025, wasting scammers an estimated 500+ days — equivalent to roughly $13 million in potential losses saved. Beyond eating up scammers’ time, another goal of the bots is to gather usable intelligence for banks and telecom companies, helping crack down on scam rings operating out of Australia, Asia, Africa, and the UK/Europe. Kaafar says the idea came to him in November 2021, when he got a scam call while having a picnic with his family in Sydney. He played along, pretending to be naive, and kept the caller on the line for 44 minutes — amusing his kids but annoying his wife — and along the way picked up a good sense of how the scam actually worked. After hanging up, he thought: if he could pull that off just for fun, imagine what it could do at scale. At the time he was teaching at Macquarie University, so he pulled in a few PhD students working on AI and cybersecurity to build an automated system together. The project later got funding from Australia’s National Intelligence Office, and in 2023 it spun out of the university to become Apate. Today the company works with most of Australia’s major banks and has expanded to banks in the UK, South Africa, and Southeast Asia. Apate isn’t the only company using AI to bait scammers — UK telecom O2 launched an “AI Granny” campaign last year that strings scammers along with cat chat, doubling as a public-awareness stunt — but Apate operates at a scale far beyond its peers. Apate currently runs 197,000 AI personas, each with distinct accents and verbal tics, trained on hundreds of hours of real recordings of humans baiting scammers.

💬 JudyAI Lab Take

The Apate case is interesting — using AI bots to pose as victims and tie up scammers’ time and resources flips AI defense toward “actively driving up the attacker’s cost,” instead of just passively detecting and blocking scam calls.

The lesson for AI builders here: in adversarial settings, AI’s value isn’t always “spot the enemy faster” — sometimes it’s “make the other side pay a much higher cost at a much lower cost to you.” With just 197,000 AI personas (each with its own accent and verbal tics) trained on real bait-the-scammer recordings, Apate handled 600,000 scam calls for a single telco in six weeks and burned over 500 days of scammers’ time. That “asymmetric attrition” mindset — using AI to scale up something a human could only ever do a handful of times — applies well beyond anti-fraud, to plenty of other defensive use cases.

Something worth chewing on next time you’re evaluating an AI application: could this thing, which only ever worked at small, manual scale, become an offense-defense advantage once AI scales it up?

📅 Original Source

🔗 Further Reading