An Australian company Apate built a network of about 350,000 AI bots that answer scam calls, chat messages and infiltrate fraud groups to keep criminals engaged and collect intelligence. The system has already gathered more than 250,000 real-time records, from scam URLs to money mule accounts and bank details. The approach turns defense from blocking attacks into draining attacker resources, which matters for banks and telecom operators facing billions of malicious calls and messages each year.
How Apate keeps scammers on the line
Apate, named after the Greek goddess of deception, has spent two years developing what founder and CEO Dali Kaafar calls perfect victims for scammers. The platform is used by banks and supported by telecom companies, which divert suspicious calls to bots instead of real subscribers. Bots also respond to text messages and enter scam chat groups online with the same task of prolonging contact. Every minute a fraudster spends with a bot is a minute not spent dialing potential victims with automated dialing tools.
The bots carry different personalities, language skills and profiles to avoid detection as machines. Some have WhatsApp accounts and some do not, some answer immediately while others hang up and promise to call back later. Personas show measured skepticism but leave openings that encourage the scammer to continue persuasion. Journalists testing a demo as fake crypto sellers failed to close an investment after six minutes, describing the exchange as responsive, natural in timing and intensely frustrating.
Long conversations are central to the model because industrial-scale scam operations depend on volume and speed. Kaafar says Apate calls regularly last more than two hours, as criminals stay hopeful of a payout. The collected material flows back as operational intelligence for defenders rather than remaining isolated anecdotes. That combination of delay and data collection separates the system from manual scambaiting, which has provided deterrence for years but could not scale against expanding fraud.
What AI disruption means for business defense
For companies that handle payments, customer calls or messaging, such automation changes the economics of fraud prevention. Instead of only filtering inbound traffic, banks and carriers can redirect it into a controlled environment that consumes criminal labor. Small firms gain access to shared intelligence on URLs, accounts and tactics without running their own security operations. Large organizations can connect bot-derived signals to fraud monitoring, transaction review and customer warnings.
The method has clear limits that buyers should examine before relying on it. Criminals initiate billions of messages and calls annually from physical scam compounds, so even a large bot swarm cannot intercept every attempt. Effectiveness depends on realism, integration with telecom routing, and sharing between police, social platforms, banks and other industries. The news alone does not prove lower losses, so vendors should be asked about coverage, false positives, data handling and measured diversion time.
The marker to watch is research on LLM-powered honeypots from ETH Zurich, where doctoral researcher Mark Vero found AI agents stayed significantly longer in realistic decoys and identified them as honeypots at a much lower rate. If banks and telecom operators publish sustained engagement times and confirmed intelligence reuse, AI-driven disruption will move from experiment to standard control. Absence of such metrics will signal that criminals adapted faster than the decoys.
