Task scam: AI-assisted fake review jobs

This case study was originally published in OpenAI’s February 2025⁠(opens in a new window) report.

Actor

We banned a cluster of ChatGPT accounts that were translating short comments between Urdu and English consistent with a “task⁠(opens in a new window)” scam, in which the victims are offered a highly-paid job, but then required to pay their own money into the system before they can access their supposed “earnings.” This activity appeared to originate in Cambodia. In these scams, victims generally lose both the “earnings” and their own money. We began investigating this activity following a tip from Meta.

Behavior

This network primarily used our models to translate short comments, consistent with chats between a scammer and their victims, between Urdu and English. We identified references to messaging on Telegram (apparently their primary platform), WhatsApp, SMS, and a range of employment forums and groups where job offers could be posted.

The scammers appeared to pose as two main types of persona. One was the “recruiter”: Their messages were a form of cold outreach, posting ads for well-paid remote work. The other, which generated a higher volume of activity, was the “mentor,” who would follow up on the “recruiter”’s outreach and teach the target how to do their job. The mentors very often referenced training sessions on Telegram.

Occasionally, the scammers also used our models to proofread the text of websites. These sites appear to have spoofed the names and appearance of luxury design, fashion, and travel brands. The “mentor” personas would claim to be recruiting on behalf of those brands, indicating that the websites were likely used to make the scam appear more credible.

Completions

Based on our limited visibility, the scammers appear to have followed a common workflow in moving their targets from engagement to scam: