OpenAI Disrupts Cambodia-Based ChatGPT Scam Network
On July 31, OpenAI disclosed that it had disrupted and banned ChatGPT accounts linked to a Cambodia-based scam network conducting investment, romance, gambling, and law-enforcement impersonation fraud. According to The Record, the network also used ChatGPT to create personas, translate target messages, generate forged-looking materials, and perform administrative work, after an investigation began with a lead from WhatsApp.
OpenAI disrupted a coordinated network of ChatGPT accounts tied to a Cambodia-based scam operation that used the service across investment, romance, gambling, and law-enforcement impersonation fraud, according to reporting by The Record and The Hacker News. The company banned accounts associated with the activity after an investigation that began with a lead from WhatsApp, the reports state.
The activity was linked to Poipet, Cambodia, according to the reports. The Record describes the location as a city publicly associated with scam compounds run by Chinese criminal gangs, while CyberPress notes that Poipet has repeatedly been linked in public reporting to online scam compounds and human-trafficking operations.
ChatGPT used across the scam workflow
According to The Record, the network used OpenAI models to build and maintain fake online personas, generate and translate messages to targets, produce promotional material, and assist with routine operations. The Hacker News reported that operators ran multiple fraud types in parallel rather than limiting activity to a single narrative.
The reported abuse included:
- •Romance and dating personas that sought to build trust before promoting fraudulent cryptocurrency or spot-gold investments.
- •Fake gambling-platform representatives offering nonexistent bonuses or winnings.
- •Law-enforcement impersonation designed to pressure targets into paying fabricated fines.
- •Images resembling passports, legal notices, stock-purchase confirmations, and gambling or trading-platform pages.
CyberPress reported that the operation may have interacted with hundreds of targets, while some conversations referenced individual losses of thousands of dollars. The Record reported that OpenAI could not calculate total financial losses associated with the activity.
Recruitment and internal administration
The use of ChatGPT extended beyond outward-facing scam content. The Record reported that some accounts drafted internal announcements, translated staff communications, and maintained records involving employee debts, salary deductions, disciplinary fines, loan repayments, immigration status, and visa matters. It also reported conversations referencing apparent detention, escape attempts, and potential criminal liability involving people trafficked into scam operations.
The Hacker News reported that promotional material included social-media advertisements for "chatter" jobs in Poipet aimed at people in Bangladesh and India. The ads reportedly promised an $800 base salary, a $100 attendance bonus, flights, accommodation, meals, visas, and work permits. The Record separately reported recruitment fliers directed at people in India with promises including free flights, accommodation, and work visas.
Developers Digest reported that OpenAI disclosed the disruption on July 31 and shared threat signals with industry partners and relevant authorities. Other coverage published August 4 and August 5 described the same investigation and account bans.
Operational lesson for AI security teams
The reported behavior illustrates how general-purpose language models can reduce friction across several stages of social engineering: content drafting, multilingual communication, persona support, visual-material generation, and internal coordination. In comparable fraud investigations, abuse detection often benefits from connecting signals across these stages rather than evaluating a single prompt, account, or scam script in isolation.
For model providers and trust-and-safety teams, the case also underscores the value of cross-platform reporting. Here, WhatsApp provided the initial lead described by the reports, while OpenAI's subsequent account action targeted activity occurring on its own service. That division of evidence and enforcement is common where messaging, social networking, payment, and model platforms each observe only part of a fraud operation.
Key Points
- 1OpenAI banned ChatGPT accounts linked to a Cambodia-based network that combined investment, romance, gambling, and impersonation fraud.
- 2Reported model use covered target-facing content, translation, persona creation, forged-looking images, recruitment material, and internal administration.
- 3Comparable multi-stage fraud cases make cross-platform threat sharing and behavioral correlation more useful than single-prompt abuse detection.
Scoring Rationale
The case offers a concrete, documented example of LLM-assisted social engineering spanning content generation, translation, persona operations, and administrative coordination. It is relevant to AI security and trust-and-safety practitioners, although the disclosed disruption occurred in late July or early August and is not a new model or platform release.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
