Yesterday OpenAI published a report showing how AI chatbots are increasingly getting pulled into geopolitical conflict. The company banned a cluster of accounts likely operated from China – accounts that used ChatGPT for an unusual campaign.
What happened
According to OpenAI, the actors used ChatGPT for covert influence operations meant to stoke opposition to building AI data centers in the US. The goal: to manipulate what is actually a legitimate debate about American AI infrastructure.
That’s notable because it shows a new angle. Earlier influence campaigns usually targeted elections or classic political division. This one targets infrastructure – specifically the data centers that power the AI boom in the US.
Why it matters
Data centers have become a political flashpoint. They consume enormous amounts of electricity and water, local communities push back against new builds, and at the same time they’re treated as a strategic resource in the race between the US and China.
The banned accounts apparently plugged straight into that already charged debate. They didn’t invent a new divide – they amplified an existing one. That makes such operations hard to spot. They disguise themselves as perfectly ordinary voices in a real controversy.
The bigger picture
It’s no coincidence that OpenAI of all companies is publishing this report. AI labs are under pressure to show they can detect and shut down misuse of their tools. Transparency reports like this have become a fixture of the industry – Anthropic regularly publishes similar analyses too.
My take: The genuinely interesting part is the choice of target. Whoever fans a debate about data centers isn’t attacking the AI models directly, but their physical foundation. No data centers, no training, no inference, no progress. That’s a clever – and unsettling – lever. And it shows how much AI infrastructure is now understood as a geopolitical battlefield. As always with reports like this: we only know the version of the company that banned the accounts. The attribution to China is plausible, but not verifiable from the outside.
Sources: Al Jazeera, OpenAI News