AI Deepfakes and Chatbots Become Tools for Industrialized Telecom Fraud
Summary
The article examines how Southeast Asian telecom-fraud networks are using real-time face-swapping, generative chatbots and integrated software to industrialize deception. A Cambodian-linked service called Haotian AI reportedly works across WhatsApp, Zoom, Microsoft Teams, TikTok, Instagram and YouTube, allowing operators to appear as another person while speaking, blinking, changing expressions and moving their hands. The report says the system fooled the Xception deepfake detector on nearly all tested samples, although visible failures can still occur when hands or facial features expose weaknesses in its three-dimensional understanding. A service introduced in April reportedly helps users pass video and bank face-authentication checks by mapping a similar-looking person to the target and reproducing requested movements such as blinking, nodding and turning the head. The software costs $1,998 per year, with customized faces priced at an additional $498, requires at least an RTX 4080-class computer, accepts TRON cryptocurrency, and the associated wallet had received more than $4 million since October 2023, according to the article. The report also describes KT, a customer-acquisition tool that connects to more than a dozen social platforms, manages multiple accounts and uses GPT- and Gemini-based features to generate messages consistent with fabricated identities. One trafficked worker reportedly used it while handling more than 100 conversations and reaching about 50,000 targets in a month. OpenAI identified a scam network in Poipet, Cambodia, that used ChatGPT to produce fake passports, legal notices, stock confirmations, gambling pages and cryptocurrency interfaces. A broader platform called 007TG reportedly sells phone numbers, social accounts, proxy IPs, SMS verification, browser-fingerprint tools and bulk messaging, while Cloud007 and SCRM Champion monitor operators, conversations, work hours, emotional responses and performance. The article cites a United Nations estimate of more than 10 deepfake-software suppliers serving Southeast Asian fraud groups on Telegram. It presents the technology as an accessible service layer assembled largely from open-source components, lowering the technical barrier while making video trust and biometric verification substantially harder to rely on.