OpenAI has shut down a scam network based in the Cambodian border town of Poipet that used ChatGPT for four different fraud schemes at once – from romance scams to fake government letters. According to the company, the investigation also turned up evidence that the operators themselves were forced to work. The decisive tip came through WhatsApp.
OpenAI Uncovers Four Parallel Scam Schemes
According to OpenAI, the perpetrators spent weeks building personal relationships with victims before shifting to the actual scam. ChatGPT reportedly helped create convincing fake identities, translate messages into multiple languages, and draft marketing copy for supposed investment opportunities. The four schemes run in parallel ranged from classic romance scams to cryptocurrency and gold investment fraud, gambling lures, and fake calls from supposed authorities designed to pressure victims further.
For the investment variant, the group used ChatGPT to build fake trading interfaces modeled on real crypto exchanges, along with forged passports, stock purchase confirmations, and official-looking letters. In its own report, OpenAI describes a recurring pattern: after small initial deposits, the scammers demanded further payments for supposed activation fees, taxes, or fines before cutting off contact for good. According to the company, the forged documents specifically targeted people who had already made smaller payments and were therefore seen as especially willing to pay. The templates used were similar enough across different languages that investigators suspect a centrally managed toolkit.
Chat Logs Reveal Forced Labor Among the Operators
According to OpenAI, internal chat logs show that part of the scam operation itself relied on forced labor. Recruitment ads, also reportedly created with ChatGPT, promised young job seekers flights, housing, food, and help with visa applications – a pattern already documented at similar scam centers across Southeast Asia.
On the ground in Poipet, internal records reportedly documented workers’ debts, wage deductions, fines, and signs of confinement and escape attempts, according to OpenAI. The scammers were themselves apparently part of a larger criminal system that has made the border town one of the region’s best-known centers of organized online fraud. Human rights organizations have described similar structures for years at comparable scam centers along the Thai-Cambodian border, where people are said to work under similar conditions. OpenAI states that its findings so far come exclusively from its own chat logs and not yet from an independent on-the-ground review.
WhatsApp Tip Triggers Months-Long Investigation
OpenAI says it received the first tip about the network via WhatsApp. The company then analyzed accounts, conversation histories, and generated content, identifying several hundred affected individuals by its own account before suspending the associated ChatGPT accounts. Before the suspension, OpenAI informed industry partners and law enforcement to enable a coordinated response.
Similar scam patterns involving AI tools are not an isolated case: just in July, a financial regulator warned about fraudsters using cloned boss voices and deepfake videos to extort urgent payments. For OpenAI, the Poipet case is, by its own account, one of the largest scam rings uncovered so far that abused its own chatbot. Human rights organizations estimate that several hundred thousand people in total are employed at comparable scam centers across the region – a figure that is independently unverified. No arrests or charges related to the case have been made public so far.
What remains open is how OpenAI intends to prevent similar networks from going undetected in the future, as long as countermeasures only kick in after an external tip like the one from WhatsApp. The case also points to a dilemma for law enforcement: shutting down accounts strips perpetrators of their tool, but it does not directly help the operators in Poipet who were themselves apparently forced to work there.
