OpenAI Busted a Cambodian Pig-Butchering Ring Using ChatGPT to Romance Victims

OpenAI has pulled back the curtain on a grimly efficient “pig butchering” operation, banning a cluster of accounts that turned ChatGPT into a tool for mass romantic deception. The network, traced back to Cambodia, didn’t just use AI to break language barriers—it used it to industrialize the entire scam pipeline, from a flirty first comment to the moment victims are convinced to dump their savings into fake crypto platforms.

The playbook was methodical. Scammers, likely native Chinese speakers, deployed the models to generate and translate comments in Japanese, Chinese, and English. They targeted men over 40 on Facebook, X, and Instagram, often those in medical fields with a public interest in golf. These weren’t random bot posts; the operators replied to real, sometimes years-old, social media posts with seemingly innocuous questions like, “What do you think about [subject]?” Their goal was simple: get a response. The profile pictures were typically stolen from real models and influencers, not AI-generated, adding a layer of stolen authenticity to the ruse.

Once a conversation started, the AI’s role shifted from public outreach to private manipulation. OpenAI’s report details how the scammers fed ChatGPT extensive fake biographies—complete with names, educational backgrounds, and a key detail that the persona “dabbled in online investment.” The model was instructed to adopt a flirty tone and was primarily used as a real-time translator, quickly pivoting conversations from public comments to secure messaging apps like LINE or WhatsApp. The excuses for the move were classic tradecraft: a professed distrust of social media or a logistical claim of being somewhere without access.

The AI-assisted workflow culminated in a textbook financial scam. After establishing an intimate relationship, the scammer would boast of a relative in finance who gave them a hot tip, showing off huge, fake profits from crypto or gold trading. They’d then urge the target to invest, guiding them into a fraudulent trading app. The final, cruelest stage saw the AI generating messages to convince victims to keep depositing more money and to silence their fears when they inevitably tried to withdraw their “profits.” Meta has separately identified related activity from a newly established scam compound in Cambodia, suggesting this is not a one-off group but an expanding criminal infrastructure that has seamlessly integrated large language models into its toolkit. The case confirms a chilling evolution: these tools don’t just help scammers scale; they make the scripts for breaking hearts and emptying bank accounts more persuasive than ever.