OpenAI has identified and banned accounts that leveraged AI to support various stages of romance scam operations. These workflows encompassed initial contact, language translation, engaging with victims, and creating lures for investment fraud. This information was originally published in OpenAI’s February 2026 report on disrupting malicious uses of AI.
Since OpenAI began reporting on scam network disruptions a year prior, numerous operations from different global regions have been taken down. These included “task” scams and investment scams, in addition to romance scams.
Key Points
- OpenAI banned accounts using AI for romance scam workflows, including outreach, translation, victim engagement, and investment-fraud lures.
- This case study was first published in OpenAI’s February 2026 report on disrupting malicious uses of AI.
- OpenAI has disrupted various attempted romance scams since its first published disruption involving a Cambodian operation using ChatGPT.
- Scams tend to follow a pattern: the ping (cold contact), the zing (generate emotion), and the sting (extract money).
- The “ping” stage involves generating content to attract targets, such as golf-themed replies or social media ads, often using ChatGPT for more engaging messages.
- The “zing” stage focuses on generating emotional responses in targets, making them more susceptible to manipulation.
- The “sting” stage involves generating content to convince targets to transfer money, with reasons varying from business investments to fictional earnings.
Context
According to OpenAI, romance scams involve perpetrators posing as potential romantic partners to convince targets they have found a love match before soliciting money. The first published scam disruption by OpenAI featured a criminal operation in Cambodia that utilized ChatGPT to create messages for a romance scam, some of which were distributed on social media. OpenAI has since disrupted various other attempted romance scams.
Why It Matters
This information details how AI models are being misused in romance scams, providing insight into the methods employed by malicious actors. For builders and deployers of AI systems, understanding these patterns is crucial for developing more robust safety measures and content moderation strategies to prevent such abuses.
What To Do
- Note the workflow stages (ping, zing, sting) described by OpenAI as common patterns in AI-enabled scams.
- Review the examples provided for each stage, such as golf-themed pings or cold-call SMS messages, to understand specific tactics.
- Consider how AI models like ChatGPT are being used to generate more engaging and less obviously non-native scam messages.
- Watch for further reports from OpenAI on disruptions of malicious AI uses to stay informed on evolving threats.