How ChatGPT Fueled a Cambodian Scam Network and OpenAI’s Counteraction

OpenAI released a concerning report revealing that it had banned a large‑scale fraud network originating from Poipet in north‑west Cambodia, a city long associated with online‑scam parks and human‑trafficking activities.

AI‑Enabled Scam Tactics

The group used ChatGPT to mass‑produce and maintain fake identities, translate and generate victim‑targeted messages, create promotional scam content, and support daily operations. Accounts were also employed for administrative tasks such as drafting internal notices, translating staff communications, and recording recruitment, immigration, work‑condition, and disciplinary matters.

Scammers combined multiple schemes, first establishing trust through romance personas, then offering “investment opportunities” involving cryptocurrency or spot gold, or posing as online gambling platforms promising false bonuses, or impersonating law‑enforcement agencies demanding fines. They generated forged passports, legal documents, stock‑purchase confirmations, and gambling‑platform screenshots.

Impact of Generative AI

Traditional romance scams relied heavily on manual effort, making it difficult to sustain consistent personas over weeks or months. Generative AI allowed perpetrators to maintain near‑perfect grammar and highly personalized tones across hundreds of concurrent conversations.

Three‑Step Harvesting Model

OpenAI distilled the scammers’ workflow into three phases: ping (initial contact via WhatsApp, Telegram, etc., using ChatGPT to translate and generate dialogue and social‑media content), zing (emotional manipulation with promises of “zero‑risk” returns, sweet talk, secrecy, and urgency), and sting (pressuring victims to pay “activation fees”, “fines”, or “reward unlock” charges and submit transfer screenshots).

Based on the scammers’ own communications, OpenAI estimates the network reached several hundred targets, with individual losses of a few thousand dollars, though total financial impact remains unverified.

Human‑Trafficking Connections

The report also highlighted that some accounts generated content related to forced labor and trafficking, such as social‑media recruitment ads for “chatters” in Poipet promising airfare, accommodation, visas, and work permits, alongside discussions of employee debt, salary deductions, loan repayments, and immigration status.

These findings align with public investigations noting at least 53 scam parks in Cambodia engaged in trafficking, forced labor, torture, and slavery, often masquerading as legitimate tech or business centers. Survivors reported confinement, forced photography for bank‑account creation, and coercion into global fraud operations.

UN and Global Context

A March 2025 UN special‑rapporteur letter warned that the Cambodian industry is a multi‑billion‑dollar criminal enterprise involving over 200,000 people, blurring lines between organized crime, scams, and trafficking.

OpenAI’s Response and Outlook

OpenAI has disabled the accounts linked to the operation, shared threat indicators with industry partners and authorities, and implemented measures to make re‑accessing its services harder for these actors. The company cautions that as long as fraud remains profitable and tools stay accessible, perpetrators will seek workarounds, suggesting stronger identity verification and payment safeguards may be more effective than platform bans alone.

OpenAI identifies two emerging trends: (1) highly diversified organized scam networks running multiple schemes simultaneously, and (2) increasingly blurred boundaries between cyber‑fraud, organized crime, and human trafficking.

Effective mitigation must target not only victim‑facing scams but also the underlying criminal organizations that profit from them.