OpenAI says its AI models have been misused across a set of international operations involving scams, cyberattacks, political influence, and employment fraud. The activity described in its threat report reaches across countries including North Korea, Russia, and Cambodia, and shows how tools like ChatGPT can be folded into very different kinds of abuse.
The clearest lesson is not that every operation looked sophisticated. Some were blunt money-making schemes. Others were more deliberate attempts to shape political conversations, gather information, or make coordinated campaigns easier to run.
How one fake job scam used ChatGPT
One scam described in the report centered on recruitment messages written with ChatGPT in multiple languages. The pitch was simple: victims were told they could earn more than five US dollars just for liking social media posts.
That promise stood out because the source article notes that it was far above typical market rates, where 1,000 likes can cost less than ten dollars. The offer was not just generous; it was designed to feel like easy money, which made it useful as a first hook.
The fraud followed a three-step pattern. OpenAI described the stages as:
- "ping": attracting victims with unrealistic offers.
- "zing": building confidence through fake testimonials and small payouts.
- "sting": pushing victims to pay supposed entrance fees or make cryptocurrency payments.
The operation was nicknamed "Wrong Number". It was discovered by chance after an OpenAI investigator received one of the first messages by SMS. The message pushed the recipient toward WhatsApp, and the victim was then moved on to a "mentor" on Telegram.
That path matters because it shows how the scam was not limited to one platform. The operation used one channel to begin contact, another to continue the conversation, and another to apply pressure through a more personal-seeming relationship.
Political influence campaigns used AI for scale
OpenAI also said it shut down a Russian influence campaign that used ChatGPT to create German-language content ahead of Germany's 2025 federal election. The campaign was called "Operation Helgoland Bite".
According to the source article, the operation ran through a Telegram channel named "Nachhall von Helgoland" and an X account with over 27,000 followers. Its posts attacked the US and NATO while promoting the right-wing populist Alternative for Germany (AfD) party.
ChatGPT was used for more than writing posts. The campaign also used it to research German opposition activists and bloggers, and to translate Russian-language propaganda into German.
OpenAI said the overall reach of the campaign was limited. Even so, the case illustrates how generative AI can support several parts of an influence operation at once: drafting, translation, research, and message adaptation for a target audience.
Campaigns with ties to China took several forms
The report also identified multiple campaigns with ties to China. They did not all use the same tactic, but each used online activity to shape, observe, or disrupt public conversation.
One campaign, "Sneer Review", posted comments about geopolitical topics on TikTok and X. Its activity aimed to discredit critics, including Pakistani activist Mahrang Baloch.
Another operation, "VAGue Focus", relied on fake media accounts to collect information on Western targets. A third campaign, "Uncle Spam", inserted contradictory statements into US political discussions, including conversations about trade policy.
"Uncle Spam" also used AI-generated profile pictures of supposed military veterans. That detail is important because profile images can make accounts appear more specific, more human, and more credible, even when the identity behind them is not real.
What connects the cases
OpenAI identified a total of ten international campaigns in which threat actors used its AI models for fraudulent, political, or criminal ends. The examples in the source article vary widely, but they share a practical pattern: AI was used to reduce the effort needed to produce language, translate messages, create personas, or manage online narratives.
In the fake job scam, ChatGPT helped generate recruitment messages across languages. In the Russian campaign, it helped produce German-language political content and translate Russian-language propaganda. In the campaigns with ties to China, the activity included comments, fake media accounts, contradictory political statements, and AI-generated profile pictures.
The report therefore points to a broad problem rather than a single type of ChatGPT scam. Misuse can look like employment fraud aimed at individuals. It can look like political content targeted at voters. It can also appear as information-gathering, comment flooding, or accounts built around synthetic identities.
For readers, the useful takeaway is to look at behavior rather than only the tool. Unrealistic pay offers, sudden requests to move across messaging apps, supposed mentors, entrance fees, cryptocurrency payments, fake testimonials, and politically charged accounts with unclear origins can all be warning signs in the kinds of activity described here.
OpenAI's findings show that AI misuse is not limited to one country, platform, or strategy. The same underlying capability can help write a scam message, translate propaganda, research targets, or generate the appearance of a real person online.