OpenAI has banned a cluster of ChatGPT accounts originating in Cambodia that were powering a classic “task scam,” a development that pulls back the curtain on how accessible AI tools are reshaping the economics of online fraud. The network, initially flagged by Meta, didn’t use the models for anything technically sophisticated—no deepfakes, no novel malware. The primary function was far more mundane and, frankly, more instructive about the current threat landscape: translating scam scripts between Urdu and English.
The operation followed a depressingly familiar playbook. Scammers, posing as “recruiters,” would blanket online employment forums with ads for remote work—too-good-to-be-true offers like $300 a day for a few hours of clicking. Once a target bit, a second persona, the “mentor,” would take over, typically steering the conversation towards Telegram. The “job” was filing fake five-star reviews for luxury brands, a task that allowed the scammers to point to spoofed websites of real design and fashion companies as proof of legitimacy. ChatGPT was used to translate and polish this back-and-forth, from the initial cold outreach to the reassurances that the work was safe.
That’s where the trap sprung. Victims would be given a “training account” where they could see fake earnings racking up. They’d be offered rare “ultimate” tasks with higher bonuses. The hook was always an “activation fee” required to withdraw those supposed earnings. OpenAI’s analysis of the generated text reveals the scammers’ psychological playbook in stark detail: if a target hesitated, the scammer would claim they too had been scammed before and then apply pressure by comparing the victim unfavorably to other, more efficient workers.
OpenAI’s report notes the conversations showed a high degree of skepticism from targets and messages from mentors asking why communication had stopped, indicating a low overall conversion rate. That’s the silver lining, but it also points to a brute-force reality. The low barrier to entry for generating convincing, multilingual text means these operations can afford to cast a very wide net, knowing that even a tiny percentage of payouts justifies the cost. It’s not a story about AI’s godlike powers; it’s a story about industrializing a con. The same translation tool a small business uses to reach new markets is, in this context, a force multiplier for a scam call center, making their fraudulent job offers sound fluent to a global pool of desperate gig workers.
