OpenAI Confirms Chinese Supporters Used ChatGPT to Plan Anti-Data Center Campaign, Platform Responsibility Debate Escalates

On June 10, 2026, OpenAI issued a statement confirming that some Chinese-affiliated actors used ChatGPT to generate anti-data center propaganda materials and tariff impact scripts. These accounts, registered with local US identities, posted content claiming that AI training and inference consume large amounts of electricity, causing residential electricity prices to rise and disrupting family life.

How Generative Models Are Used in Influence Operations

Large language models like ChatGPT, trained on massive text corpora, can rapidly produce coherent social media posts, comments, and video scripts. Actors input keywords such as "data center electricity consumption" or "household electricity price increase," and the model outputs emotionally charged short texts, which are then manually fine-tuned and published. This process compresses copywriting work that would otherwise take hours into minutes, reducing coordination costs.

OpenAI system records show that related accounts generated similar narrative frameworks in large batches within a short period, including specific phrases such as "AI data centers hogging local electricity" and "ordinary households paying hundreds more in electricity bills each month." These contents were targeted and optimized for local election cycles and energy policy discussions in the United States.

Platform Detection and Termination Mechanisms

OpenAI identified anomalies through behavior pattern analysis: accounts were registered in concentrated time periods, content topics highly overlapped, and interaction networks exhibited a star-shaped structure. Before June 10, the system had already terminated dozens of related accounts. Termination criteria included IP address clustering, duplicate device fingerprints, and semantic similarity thresholds of generated texts.

This process relies on metadata output by the model itself and downstream classifiers. The classifiers are trained on data from historically known influence operations, focusing on capturing the combination of "fake local identities + policy issues + emotional incitement."

Impact on AI Infrastructure

Data center site selection is directly constrained by power supply and local regulations. Since 2025, several US states have delayed or reduced AI computing projects due to similar public opinion pressure.

From a technical perspective, the energy consumption of AI training and inference mainly comes from GPU clusters and cooling systems. A single large-scale training run can consume millions of kilowatt-hours of electricity, but the actual increase in residential electricity bills depends on the grid structure and local pricing mechanisms. OpenAI has not disclosed specific energy consumption figures, only emphasizing that its termination actions were based on account behavior rather than content authenticity.

The Intersection of Geopolitics and Platform Responsibility

The core of platform responsibility lies in balancing content moderation and freedom of speech. OpenAI's approach is ex-post banning rather than ex-ante filtering.