we_are_coded.by CODE · The world, decoded
БГ

OpenAI · Incidents

OpenAI shut down a scam factory in Cambodia that used ChatGPT for both the scams and the paperwork

The company blocked a coordinated network of accounts, most likely operating from the area of Poipet - a city linked to online scam compounds. The schemes: romance, investment, gambling and fake police. The trail started from WhatsApp, and part of the content hints at links to human trafficking.

Source
OpenAI
Published
The event
Checked on
Responsible editor
Cvetelin Ivanov
How we work
MethodCorrections

It wasn't the scams that stopped me in this report. The paperwork. The scam group used ChatGPT the way you'd use an office suite - internal messages, translations between staff, documents for hiring and discipline. The crime is organized like a company, and it keeps its own records.

Fact

The facts: on 31 July OpenAI reported it had blocked a coordinated network of ChatGPT accounts which, per the company's data, very likely operated from Cambodia, in or around the city of Poipet - a place that public investigations have repeatedly linked to online scam compounds. The network ran several schemes at once: romance contacts that, once trust was built, turned toward 'investments' in cryptocurrency and gold; fake bonuses from gambling platforms; and posing as police demanding payment of 'fines'. The models were used to create and maintain fake personas, translate messages to victims, produce promotional content, and handle day-to-day admin. The investigation started from a tip by the WhatsApp team; OpenAI says it has shared indicators with industry partners and relevant authorities. Some of the generated content hints at links to human trafficking and forced criminality - consistent with public reports on the region. Source: OpenAI, 31.07.2026. There is no CVE: this is a scam operation shut down at the account level, not a software flaw.

There are two layers here, and the second is darker than the first. The first is technical: language used to be the natural barrier against these factories. A message in broken Bulgarian gives itself away from a mile off; a message run through a model doesn't. The second layer is human: OpenAI explicitly notes that the people writing the scams may themselves be victims - lured with a job promise and locked into debt. The factory scams outward, and inward.

Broken language used to be the cheapest scam detector. It's gone now.

If this sounds familiar: a stranger who quickly gets close and then brings up an investment; a 'profit' that first requires paying a fee; a 'police officer' collecting fines in chat. The plots have been the same for years. Only the spelling changed. A government institution doesn't collect money through a messaging app. That's it.

The visual is generated code-art, no third-party images.