AI Is Getting Really Good at Messing With Cybercriminals — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Apate has spent two years building AI agents that divert phone scammers onto calls designed to keep conversations going indefinitely without ever falling for the con.
- Apate's platform, used by banks and supported by telecom providers, runs roughly 350,000 bots across voice, chat, and SMS and has collected more than 250,000 pieces of fraudster intelligence in real time.
- Dali Kaafar, Apate's founder and CEO, says individual scam calls regularly stretch beyond two hours and that each minute a scammer spends on a bot protects 'hundreds, if not thousands' of potential victims.
- Mark Vero and colleagues at ETH Zurich found in recent research that LLM-powered honeypots kept attacking AI agents engaged 'significantly longer' than traditional decoys with predictable behavior.
- Open-source honeypot providers have increasingly incorporated large language models to make decoy systems appear more realistic and harder for attackers to flag.
- Even with Apate's 350,000-bot swarm, the piece notes cybercriminals still initiate billions of calls and messages a year, and law enforcement efforts have not stopped the expansion of digital scamming overall.
Why it matters: Apate's 350,000 bots stretch individual scam calls past two hours, and Kaafar estimates each minute a scammer wastes with a bot protects hundreds or thousands of potential victims from the same operation. The ETH Zurich honeypot finding adds a second front: defenders can now use LLMs not just to study attackers but to keep them tied up longer than legacy decoys, shifting the per-attack cost toward cybercriminals.
Ask SkimNews
