OpenAI has shut down a coordinated network of ChatGPT accounts operating from Poipet, Cambodia, a city with well-documented connections to scam compounds and human trafficking, according to a disclosure published by the company. The network used the company's generative AI chatbot to run investment, romance, gambling, and law enforcement impersonation scams across Southeast Asia. OpenAI banned the cluster of accounts, which it says likely originated from Southeast Asia, after investigating the operation in partnership with Meta-owned WhatsApp.
The scammers deployed ChatGPT across the full lifecycle of their operations, the company said. They used OpenAI's models to build and maintain fake online identities, compose and translate messages to scam targets, and produce promotional materials for their fraudulent schemes. The promotional content featured social media ads recruiting "chatter" workers in Poipet, specifically aimed at users in Bangladesh and India, promising a base salary of $800 with a $100 bonus for full attendance, plus flight tickets, free housing, meals, one-year Cambodia visas, and work permits. A portion of the accounts also used the AI tool for back-office tasks, including writing internal announcements, translating staff communications, and recording employee debts, salary deductions, fines, loan repayments, visa overstays, work permits, immigration status, and recruitment incentives.
The scammers ran multiple fraud schemes simultaneously, blending tactics to maximize their success rate. They created dating personas to establish trust before luring victims into fake investment opportunities involving cryptocurrencies and spot gold trading, OpenAI reported. Other operatives engaged in prolonged romantic exchanges using synthetic identities or posed as representatives of online gambling platforms offering fictitious bonuses and winnings. Still others impersonated law enforcement officials to create a false sense of urgency, telling targets they needed to pay fines for alleged serious criminal offenses. The network generated images of forged documents—passports, legal notices, stock-purchase confirmations, and gambling platform interfaces—to support these schemes. OpenAI noted that "organized criminal groups rarely restrict themselves to a single type of scam," instead deploying "whatever narratives, personas, and tactics they think will be most effective to deceive victims."
The scammers followed a three-step attack pattern OpenAI calls "ping-zing-sting": initial outreach on messaging platforms like WhatsApp and Telegram, a trust-building phase, and finally instructing victims to make deposits, pay activation fees, or settle non-existent fines. Victims received fake screenshots of transfers or account information as supposed proof of payment. Some accounts in the network created content consistent with human trafficking and forced labor tied to organized crime groups in East Asia—operations known to recruit people with false job promises before confiscating their passports and forcing them to work in slave-like conditions. The full financial toll remains unknown, but based on the scammers' own internal communications, the operation may have contacted hundreds of targets across multiple scam types, with individual victims losing thousands of dollars, though OpenAI cautioned it couldn't independently verify those claims. The disclosure highlights how threat actors are building AI-augmented offensive capabilities that let them dramatically accelerate the speed and scale of their campaigns, even as frontier models themselves have been observed targeting real systems and individuals during Capture-the-Flag-style cyber evaluations. For organizations deploying generative AI, the episode underscores the challenge of distinguishing legitimate automation from criminal exploitation when the same tools serve both ends equally well.