One Wrong-Number Text Blew Its Cover: OpenAI Disrupts a Cambodia-Linked Task Scam Network

Hi, it's Shiichan! Today's story made me sit up a little straighter, so let me walk you through an OpenAI report on disrupting a scam network.

OpenAI News openai.com

What was announced?

OpenAI's News reported that it banned ChatGPT accounts linked to a network running task scams. The network mass-produced recruitment-style messages offering high pay for trivial tasks like liking social media posts, translated into English, Spanish, Swahili, Kinyarwanda, German, and Haitian Creole.

OpenAI named the operation "Wrong Number," after an SMS the network actually sent out. The operation appeared highly centralized, and OpenAI reports it likely originated from Cambodia.

Why it matters

Most of what this network actually used ChatGPT for was translation: converting conversational messages between Chinese and languages like English, Spanish, Kinyarwanda, Swahili, German, and Haitian Creole, then translating replies back into Chinese.

In other words, AI translation let the scam operators reach victims across multiple countries and languages without needing to speak those languages themselves. On the flip side, OpenAI says it used its own AI-powered translation tools to investigate and disrupt the campaign quickly across all those languages too. It's a good reminder that AI's convenience cuts both ways, for attackers and defenders alike.

What changes

By banning the accounts tied to this network, OpenAI stopped it from generating new recruitment messages and translated conversations through ChatGPT, at least for now. That said, the network operated across multiple messaging platforms, including SMS, WhatsApp, Telegram, and BonChat, and the report notes that the early template messages on those channels didn't appear to be AI-generated. So it's worth being honest that banning ChatGPT accounts alone doesn't mean the whole scam operation stopped.

Dive Deep

OpenAI describes the network's workflow as a three-stage pattern: "the ping, the zing, and the sting."

• The ping (cold contact): Reaching out to strangers cold, offering unusually high pay for simple tasks or big returns on stock investments
• The zing (generate enthusiasm): Translating conversations between the "operator" and "employees," mixing task logistics with motivational messages about earnings and bonuses
• The sting (extracting money): Pressuring targets to pay a "deposit," buy cryptocurrency, or cover "handling fees" to unlock bigger rewards

The whole thing kicked off with an SMS sent to seemingly random UK mobile numbers, one of which happened to belong to an OpenAI investigator, which is where the "Wrong Number" name comes from. From there, recipients were routed to WhatsApp, and then to a "mentor" on Telegram.

The messages repeatedly referenced fake employer names like Hyesung Advertising and Lightning Shared Scooter Co (LSSC), which outside reporting has already flagged as alleged task-scheme brands. OpenAI didn't independently verify what those companies actually are, but its investigation found that messages claiming to come from them were generated by Chinese-speaking ChatGPT users, likely operating from Cambodia.

The claimed industries ranged from stock trading to scooter rentals to selling social media likes, and one pricing detail stood out as a red flag: the network offered more than $5 for a single TikTok like, while real online marketplaces typically sell 1,000 likes for under $10.

Specific examples in the report include a cold-call message sent to eight phone numbers at once (none of which were in the recipient's contacts), and a Telegram message instructing a target to buy £20 worth of cryptocurrency and send it to an unnamed merchant. Outside reporting suggests these operations charged new recruits steep joining fees and used part of that money to keep existing "employees" paid just enough to stay engaged, a structure that's characteristic of task scams.

It's hard to pin down the network's true reach given limited visibility, but off-platform reports and "employees" demanding refunds suggest that at least some people actually paid money. OpenAI also observed genuine users defending these companies on social media, which suggests some real people got drawn in without realizing it was a scam.

Wrap-up

• OpenAI banned ChatGPT accounts linked to a task scam network likely operating from Cambodia
• It mass-produced high-pay recruitment messages and translated conversations in English, Spanish, Swahili, Kinyarwanda, German, and Haitian Creole
• The workflow followed a three-stage pattern: the ping (cold contact), the zing (build enthusiasm), and the sting (extract money)
• It routed victims through SMS, WhatsApp, and Telegram, using fake employer names like Hyesung Advertising and LSSC
• Some victims appear to have actually paid money, and some real users unknowingly defended the scheme on social media

If you're curious how AI gets misused, and how it gets caught, this report is worth a read!