Scam operations: Online fraud networks

This case study was originally published in OpenAI’s October 2025⁠(opens in a new window) report.

Networks likely originating in Cambodia, Myanmar, and Nigeria abusing ChatGPT in what appear to be attempts to defraud people online.

We are dedicated to identifying, preventing and disrupting attempts to abuse our models for harmful ends. Among those abuses are scams, where we detect and ban malicious actors attempting to misuse our models to deceive and defraud people. In the last three months, we’ve disrupted scam networks that likely originated in Cambodia, Myanmar and Nigeria. In an indication of the role AI can play in combating fraud, in the process of investigating these patterns of abuse, we’ve also seen many people using our models to help them identify scams. This report discusses both use cases.

Abuse of our models to support scams ranges from lone actors attempting fraud to scaled and persistent operations likely linked to organized crime groups. Regardless of their origins and precise tactics, the scam-related activity we’ve disrupted typically follows a common pattern, which we think of as the ping (cold outreach), the zing (trying to generate enthusiasm or panic), and the sting (extracting money or valuable information).

These scammers start out by scattering content, whether AI-generated or not, across messaging services and the internet, including by running social media ads. They then attempt to inspire anyone who replies with either enthusiasm for a lucrative opportunity or fear of some imminent financial loss, and leverage that emotion to convince the target to hand over money or sensitive information.

Example of an OpenAI investigator pasting a screenshot of a scam SMS message they received into ChatGPT and successfully using the model to identify it as a scam. In this instance, the threat actors attempted to impersonate TikTok recruiters.