OpenAI has cracked down on a sophisticated network of romance scams that leveraged its models to automate deception. In its February 2026 report, the company detailed how it banned accounts using ChatGPT to manage every stage of these fraudulent operations, from initial outreach to final financial extraction. This isn’t just about spam bots; it’s about criminal enterprises using AI to make their lies more convincing and their targeting more precise. The scale of the disruption highlights a growing arms race between platform security and bad actors who view large language models as force multipliers for fraud.
The scammers followed a distinct three-step playbook: ping, zing, and sting. First, they used AI to craft cold-contact messages that felt less like generic spam and more like genuine interest. One notable campaign targeted American men in their 40s working in medical fields, replying to their golf-related social media posts with personalized comments. By removing the telltale signs of non-native English speakers, the AI helped scammers bypass the initial skepticism that usually protects potential victims. This level of customization allowed them to cast a wider net while maintaining the illusion of authenticity.
Once contact was made, the AI shifted to generating emotional hooks. The goal was to build rapid intimacy or excitement, making victims more susceptible to manipulation. In romance scenarios, this meant crafting loving messages that accelerated the relationship timeline. In other cases, it involved creating urgency around fake investment opportunities or legal threats. The technology didn’t just write words; it engineered psychological pressure. When the trust was established, the final stage involved asking for money, often framed as an investment in a shared future or a solution to a sudden crisis.
What makes this trend particularly worrying is the shift from scattergun approaches to precision targeting. While older scams relied on volume, these AI-enabled operations used data to find specific vulnerabilities. OpenAI’s intervention stops the tool, but not the intent. As distribution channels evolve from email to encrypted apps and social ads, the core tactic remains the same: exploit human loneliness and greed with machine-generated empathy. The real challenge now isn’t just detecting the scam, but identifying the subtle linguistic shifts that signal a machine is doing the courting.
