OpenAI disrupts crime ring using ChatGPT for scams

OpenAI recently dismantled a coordinated network of ChatGPT accounts originating from Cambodia, specifically centered around the city of Poipet in Banteay Meanchey province. Acting on an initial tip from security peers at WhatsApp, the artificial intelligence developer banned the accounts after discovering they were leveraging the large language model to run multifaceted fraudulent schemes. The criminal group utilized the AI to generate fake online personas, translate messages, and draft promotional materials for romance, gambling, investment, and law enforcement impersonation scams.

The threat actors targeted hundreds of victims, with internal communications indicating individual losses of thousands of dollars. To execute their operations, the scammers relied on ChatGPT to facilitate a three-step process: the ping for initial contact, the zing to build emotional trust, and the sting to extract money. Beyond victim-facing deception on platforms like WhatsApp and Telegram, the network also used the AI for internal administrative tasks. This included translating staff communications and documenting employee debts, salary deductions, and disciplinary fines, which pointed to broader human trafficking and forced labor operations within the scam compounds.

For AI safety practitioners and trust and safety teams, this disruption highlights the increasingly blurred lines between digital abuse and physical organized crime. It demonstrates that malicious actors do not limit themselves to a single vector, but rather deploy highly diversified, multi-layered campaigns. Practitioners must look beyond isolated instances of spam to identify broader behavioral patterns and coordinate threat intelligence sharing across platforms. As OpenAI did by sharing signals with industry partners, combating these sophisticated networks requires collaborative, cross-platform defense strategies that target both the technological abuse and the underlying criminal infrastructure.