OpenAI’s threat intelligence team published a report detailing two influence operations that leveraged its chatbot to generate synthetic content aimed at shaping U.S. public opinion on AI infrastructure. The first cluster, dubbed “Data Center Bandwagon,” consisted of accounts that asked ChatGPT to produce English-language talking points and comic strips framing data centers as drivers of electricity price spikes. Operators posed as concerned Americans across social media, posting AI-generated material that cited legitimate news about capacity auctions and rising energy costs—a real engineering challenge as data center loads strain regional grids. According to Bloomberg, monthly electricity prices near some facilities have surged up to 267% over five years.

The operators uploaded strategy documents to ChatGPT outlining their objectives: sway public opinion, evade detection, and even generate insults targeting Chinese dissidents. OpenAI assessed the group as a private Chinese company’s social media team working for local government clients. A second cluster focused on U.S. tariff policies and tech criticism, generating multilingual comments in English, Italian, Japanese, and traditional Chinese, while explicitly instructing the model to exclude images of President Xi Jinping.

From an engineering standpoint, the campaigns are notable for their low sophistication—they relied on direct prompt injection rather than API automation or model fine-tuning. OpenAI notes the operations failed to achieve meaningful engagement, partly because they amplified existing, organic debates. The report raises a practical question for platform engineers: how do you distinguish between coordinated inauthentic amplification and legitimate grassroots concern over data center externalities? As for why operators chose ChatGPT over domestic alternatives like DeepSeek, OpenAI’s report offers no conclusion—a reminder that attribution in adversarial machine learning remains an open problem.

Source: Engadget