OpenAI Case Study on AI-Enabled Romance Scams

OpenAI has identified a recurring three-stage workflow—the "ping," the "zing," and the "sting"—that scam networks use to leverage AI for romance, investment, and task-based fraud. This framework describes how threat actors use Large Language Models (LLMs) to increase the engagement and persuasiveness of their outreach to defraud victims.

The AI-Enabled Scam Workflow

Malicious actors use AI to refine the psychological manipulation of their targets across three distinct phases:

1. The Ping (Cold Contact)

The ping is the initial contact designed to attract a target's attention. Threat actors use ChatGPT to generate messages that are more engaging and appear less like non-native speech than traditional scams, making them more convincing.

  • Targeting Methods: Distribution varies from scattergun approaches to precision targeting. Examples include replying to social media posts about golf to target American men in their 40s in the medical profession, using cold-call SMS, fake recruitment messages, or social media ads.
  • Example: In a "pig butchering" scam reported by OpenAI in 2025, the "ping" involved tailoring messages to the specific interests of the target, such as golf, to establish initial rapport.

2. The Zing (Generate Emotion)

The zing phase focuses on triggering strong emotions to make the target easier to manipulate. Depending on the scam type, the goal is to create attraction, excitement, or fear.

  • Romance Scams: The objective is to make the target feel a romantic attraction to the scammer.
  • Investment/Task Scams: The goal is to create excitement about a lucrative deal, a fear of missing out (FOMO), or alarm regarding an alleged legal risk or unpaid bill.

3. The Sting (Extract Money)

The sting is the final phase where the scammer uses AI-generated content to convince the target to transfer funds. The justification for the request varies by scam type:

  • Romance Scams: Requests may involve investing in a partner's business or providing money to cover a financial crisis.
  • Task Scams: Targets are told they must pay money to access fictional earnings they believe they have already earned.
  • Investment Scams: Targets are part of a fraudulent scheme involving non-existent investment companies.

Distribution Networks and Targeting

While AI enhances the content of the messages, the distribution network remains a critical component of the operation's success. OpenAI notes that the distribution method—whether scattershot or targeted—significantly impacts the ability to exploit targets.

  • Distribution Channels: Current operations use SMS, encrypted messaging apps, social media posts, and online ads. This contrasts with pre-AI era methods such as email, phone calls, letters, and telegrams.
  • Targeting Strategies: Some operations employ precision targeting. For instance, one romance scam utilized social media ads to specifically target young men in Indonesia, while celebrity scams target fan groups of famous individuals.

Sources