OpenAI has terminated a series of ChatGPT accounts connected to a Cambodian criminal organization that used the chatbot to facilitate cryptocurrency investment fraud. The case adds to a growing body of evidence showing how organized crime groups are exploiting generative AI tools alongside digital asset payment systems.

The shutdown comes amid a rise in digital asset–linked scams, prompting exchanges, token developers, and AI companies to strengthen preventive measures. So-called "pig-butchering" scams — a term derived from the Chinese phrase "shā zhū pán," where fraudsters build romantic or investment relationships with victims before persuading them to transfer cryptocurrency — remain widespread across Southeast Asia, increasingly aided by AI-generated content.

How Operators Exploited ChatGPT

According to OpenAI's report, the investigation was triggered by a tip-off from WhatsApp. The activity was traced back to accounts allegedly based in Poipet, a Cambodian border city long known to international investigators for its association with large-scale scam operations, many of which are believed to have expanded from infrastructure built for the region's casino and online gambling industry during the pandemic.

OpenAI stated that the operators used ChatGPT to create fake identities, translate conversations into multiple languages, and compose messages designed to build trust with targets. The chatbot was also used to generate marketing materials promoting fraudulent investment schemes.

Cryptocurrency was central to the operation. Victims were first drawn into romantic or platonic conversations, then directed to fake cryptocurrency and spot-gold trading platforms. To reinforce the illusion of legitimacy, the perpetrators used ChatGPT to fabricate fake passports, legal notices, stock purchase confirmations, and phony crypto trading dashboards.

OpenAI described the fraud as unfolding in three stages: criminals first convinced victims to send cryptocurrency, then demanded additional payments under the pretext of activation fees, taxes, or regulatory fines.

Evidence of Labor Trafficking

OpenAI's investigation also uncovered indicators of labor trafficking. Some operators used ChatGPT to design recruitment advertisements for "chatters" in Poipet, offering incentives such as free travel tickets, lodging, food, and visa assistance. Other communications involved creating spreadsheets to track employees' debts, salary deductions, fines for misconduct, immigration status, detention, and attempted escape.

These findings echo research by Elliptic, which found that many individuals behind romance and investment scams are themselves victims of human trafficking, forced to work in organized crime compounds across Southeast Asia. In August 2023, the United Nations Office of the High Commissioner for Human Rights estimated that at least 120,000 people in Myanmar and approximately 100,000 in Cambodia were being held in compounds and forced to carry out online scams, with victims originating from across the region and beyond.

Before shutting down the accounts, OpenAI shared its findings with industry stakeholders and law enforcement. Based on internal communications, the company estimates the network contacted hundreds of victims, with combined losses potentially reaching thousands of dollars.

Illicit Crypto Flows Reach Record Levels

The scale of crypto crime adds broader context to OpenAI's action. According to TRM Labs' Crypto Crime Report released on January 28, illicit cryptocurrency flows were estimated at $158 billion in 2025 — an increase of nearly 145% from the previous year and the largest figure in five years. However, the share of illicit activity within total on-chain volume declined from 1.3% in 2024 to 1.2% in 2025, reflecting wider adoption of cryptocurrency.

The report also found that stablecoins accounted for approximately 84% of verified fraudulent inflows, cementing their role as the dominant payment vehicle for organized scam operations. Law enforcement agencies worldwide have responded with intensified enforcement, including the U.S. Department of Justice's National Cryptocurrency Enforcement Team and cross-border coordination through the Financial Action Task Force, though the borderless and pseudonymous nature of crypto payments continues to complicate asset recovery.

AI Companies as a Chokepoint

OpenAI's intervention is notable because generative AI can now perform the most labor-intensive component of scams: building trust with victims. A study by researchers from Amrita Vishwa Vidyapeetham, Ca' Foscari University of Venice, the University of Melbourne, and Ben-Gurion University of the Negev demonstrated that an AI-enabled Claude agent outperformed a human scammer in conducting a pig-butchering scam. After one week of messaging, 46% of participants used the software recommended by the AI agent, compared with only 18% when interacting with a human operator. Participants also rated the AI as more reliable.

The researchers interviewed 145 individuals previously involved in scamming operations, including survivors of trafficking from facilities in Cambodia, Myanmar, and Laos. Their findings indicate that AI's primary role in scam operations is automating the trust-building phase, and that restricting access to sophisticated AI models can disrupt scams before victims are persuaded to part with their money.

OpenAI and other leading AI labs have increasingly published threat intelligence reports documenting how their models are exploited by state actors, cybercriminals, and influence networks. As organized crime syndicates increasingly leverage generative AI and cryptocurrency payment rails, preventing malicious use of advanced AI tools may become as critical to cybersecurity efforts as blocking illicit wallets or dismantling money laundering networks.