OpenAI Busted a ChatGPT-Powered "Cheerleading" Network Around Rwanda's Election!

Hey, it's Shiichan! I came across a story about AI being used in a not-so-great way, so let's dig into it together. It's less "exciting" and more "hmm...", but let's take a look!

OpenAI News openai.com

What was announced?

OpenAI's News page reported that it banned a set of ChatGPT accounts originating in Rwanda. People inside the country had been using ChatGPT to generate short comments supporting the ruling Rwandan Patriotic Front (RPF) ahead of the country's elections, then posting them on X (formerly Twitter). Researchers at Clemson University first spotted the activity, and their report led OpenAI to investigate and eventually suspend the accounts.

Why it matters

With generative AI, you can churn out huge volumes of election-flavored "cheerleading" comments in very little time. That's basically what happened here: similar-sounding comments were mass-produced in three languages — English, French, and Kinyarwanda. Some identical tweets even went out from multiple accounts almost simultaneously, which doesn't look like the work of people carefully writing each post by hand. Elections are exactly the kind of high-stakes moment where this "volume over substance" style of influence operation can matter, so how well platforms can detect it is a genuinely important question.

What changes

OpenAI first banned the initial batch of accounts tied to this activity. But it didn't stop there — it kept monitoring for newly created accounts that picked up similar patterns after the ban, and suspended those too as they appeared. In other words, this wasn't a one-and-done fix; it's an ongoing, whack-a-mole style enforcement effort.

Dive Deep

The posts used hashtags like #RPFOnTop, #PKNiWowe, and #ToraKagame24, some of which briefly made it into X's top trends in Rwanda. That said, these hashtags were also used by ordinary users, so the trending wasn't solely the network's doing.

The scale was substantial: some accounts posted hundreds of tweets per hour, and the network produced at least thousands of tweets overall. Yet individual posts barely got any engagement — replies, likes, and shares stayed in the single digits, and many posts got none at all. That gap between huge output and near-zero engagement is a textbook signature of a comment-spamming network.

One interesting detection detail: when you try to generate huge volumes of short text around one narrow theme, there are only so many synonyms and phrasings available. That limited variation meant a lot of the posts ended up looking very similar in wording and structure, which is exactly what helped reveal the pattern as AI-generated.

Wrap-up

• OpenAI banned a set of Rwanda-origin ChatGPT accounts that mass-posted short pro-RPF comments on X ahead of the country's elections
• The activity was first flagged by researchers at Clemson University
• Similar-sounding tweets were produced in English, French, and Kinyarwanda, with identical content posted from multiple accounts almost simultaneously
• Despite huge posting volume, engagement (replies, likes, shares) stayed in the single digits or at zero — a classic spam-network pattern
• OpenAI kept banning newly created copycat accounts that appeared after the initial takedown

If you care about platform integrity, AI misuse prevention, or election-related information security, this is a quiet but worthwhile case to know about!