OpenAI's February 2025 threat report says the company banned ChatGPT accounts linked to a Cambodia-origin romance and investment scam network. The operators used the model to translate and generate social-media comments in Japanese, Chinese, and English, while OpenAI assessed that some profile images were copied from real models or influencers rather than generated by AI.
OpenAI's February 2025 threat report says the company banned a cluster of ChatGPT accounts used to generate and translate content for a suspected Cambodia-origin romance and investment scam network. The company described the activity as consistent with "pig butchering," a fraud pattern in which operators cultivate trust before steering victims toward fraudulent investments.
OpenAI reported that the operators' primary language appeared to be Chinese and that they used ChatGPT to create short comments in Japanese, Chinese, and English. The comments were posted publicly on Facebook, X, and Instagram through accounts that typically used profile pictures of young women. OpenAI assessed that at least some of those images had been copied from real models or influencers rather than generated by AI.
Reported targeting pattern
According to OpenAI, the accounts commonly replied to posts made by real people, including posts that were months or years old. The network frequently targeted men whose public profiles suggested they were over 40, often in medical professions, and it regularly engaged with golf-related posts. OpenAI described occasional comments about politics and current events as well.
The report does not establish that the actors used ChatGPT to execute transactions, access victims' financial accounts, or generate the profile images. Its documented role was language generation and translation for social-media outreach.
Abuse operations and detection
The case illustrates how general-purpose language models can reduce the effort required to localize scam outreach across languages and platforms. In comparable fraud operations, reusable prompt workflows and rapid text generation can increase message volume while preserving a superficially conversational tone, making platform-level behavioral detection as important as text-level moderation.
For security and trust-and-safety teams, the reported indicators are operationally relevant: recently created accounts, reused or impersonated profile images, replies to old posts, narrow demographic targeting, and repeated engagement around particular interests. Those signals are not independently dispositive, but combined they can help investigators distinguish coordinated outreach from ordinary social-media activity. OpenAI's report documents an account-enforcement action from February 2025, not a new model capability or an August 2026 disruption. BioCatch discussed the same case in March 2025, while Reuters later documented the broader use of ChatGPT inside Southeast Asian scam compounds. The original event should therefore be read as an archived enforcement case with continuing relevance to fraud detection.
Key Points #
- 1OpenAI's February 2025 threat report documented account bans tied to Cambodia-origin romance and investment scam outreach generated or translated through ChatGPT.
- 2The reported operation combined multilingual text generation with impersonation-style social profiles and demographic targeting across social platforms.
- 3OpenAI documented language generation and translation, not AI execution of transactions or creation of the profile images.
Scoring Rationale #
The archived case provides concrete evidence of language-model assistance in multilingual social-engineering workflows and useful detection signals for trust-and-safety practitioners. Its underlying enforcement action dates to February 2025 and does not describe a new technical capability, limiting its immediate impact.
Sources #
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.