OpenAI removes ChatGPT accounts linked to Poipet scam operation
OpenAI removes accounts tied to a Poipet scam operation
OpenAI has banned a coordinated network of ChatGPT accounts likely operating from Poipet, Cambodia, after finding that they supported investment, romance, gambling and law-enforcement impersonation scams. The company said the cluster used its models to create fake online personas, write and translate messages to targets, produce promotional material and assist with routine operations.
OpenAI investigated the activity with Meta-owned WhatsApp. It described the network as originating in Southeast Asia and connected it to Poipet, a city associated in the past with scam compounds and human trafficking. The full scale of losses is unknown, although OpenAI said communications referenced individual victims losing thousands of dollars and suggested the operation may have contacted hundreds of targets.
Multiple fraud narratives used in parallel
The actors did not rely on one fraud format. They used dating personas to establish trust before steering people towards supposed cryptocurrency and spot-gold investments. Other interactions involved long romantic conversations under synthetic identities, or personas claiming to represent online gambling services offering fictitious bonuses and winnings.
Another part of the cluster impersonated law-enforcement agencies, telling targets that they had committed serious offences and needed to pay fines. OpenAI said the operation created fake dating profiles, invented investment experts and fraudulent official identities. It also generated images of forged passports, legal notices, stock-purchase confirmations and gambling-platform interfaces.
The fraud sequence followed a three-stage model OpenAI calls ping-zing-sting: initial contact through messaging services such as WhatsApp and Telegram, a period of trust building, then a request for a deposit, activation fee or non-existent fine. Targets could subsequently receive fake screenshots of transfers or account details intended to act as proof of payment.
Operational use extended beyond victim messaging
ChatGPT use also extended to administration. A subset of accounts drafted internal announcements, translated staff communications and recorded employee debts, salary deductions, fines, loan repayments, visa overstays, work permits, immigration status and recruitment incentives. Advertisements for “chatter” jobs in Poipet targeted people in Bangladesh and India, offering an $800 base salary, a $100 full-attendance bonus, flights, accommodation, meals, one-year Cambodia visas and work permits.
Some content was consistent with human trafficking and forced labour associated with organised crime groups in East Asia. OpenAI noted that recruiters may make false promises of well-paid jobs, confiscate passports and compel recruits to work in slave-like conditions.
For businesses, the case reinforces that fraud controls should not focus on a single narrative. Payment verification, escalation paths for urgent requests and awareness training need to account for romance, investment, gambling and false-authority lures, including risks highlighted by AI-agent workspace phishing risks in AI-agent workspace phishing, while teams preserve suspicious messages and payment evidence for review.