The propaganda machines just got their hands on the propaganda machine.
The Summary
- OpenAI published evidence of PRC-linked influence operations using AI tools to manipulate U.S. tech policy debates, targeting narratives around data centers, tariffs, and ChatGPT itself.
- The campaigns blend AI-generated content with coordinated posting to shape perception on platforms where policy gets debated before it gets written.
- This isn't future threat modeling. It's happening now, at scale, with your tools.
The Signal
OpenAI's Threat Intelligence team caught PRC-linked networks running coordinated influence campaigns using AI-generated text, images, and personas across social platforms. The targets: U.S. debates on AI regulation, data center energy consumption, semiconductor tariffs, and misinformation about ChatGPT's capabilities and origins. Not speculative. Detected, documented, disrupted.
The operations used ChatGPT and similar models to generate persuasive posts at volume, then deployed them through networks of inauthentic accounts. The content quality was high enough to blend in. Not the broken-English spam of 2016. Grammatically clean arguments that sound like they came from concerned citizens, industry insiders, or policy wonks.
"These aren't bots screaming into the void. They're scaled persuasion engines wearing the mask of organic debate."
What makes this different from past influence ops:
- Production speed: AI collapses the time from strategy to scaled content from weeks to hours.
- Linguistic quality: No more tells. The grammar is native, the framing sophisticated.
- Adaptive targeting: Campaigns shift messaging based on what's trending in real policy discussions.
The data center angle is particularly telling. PRC-linked accounts pushed narratives that U.S. data centers are environmental disasters while Chinese facilities are models of efficiency. The goal: slow U.S. AI infrastructure buildout while China sprints. They're not just stealing research. They're poisoning the regulatory environment.
OpenAI disrupted the networks by banning accounts and sharing IOCs (indicators of compromise) with platforms and researchers. But here's the structural problem: the same tools powering legitimate users also power coordinated inauthentic behavior at scale. Every capability you ship is dual-use by default.
The Implication
If you're building AI tools, you're now in the geopolitical targeting business whether you wanted to be or not. Your models will be weaponized for influence operations. Your mitigations need to assume adversarial nation-state users, not just spammers and scammers.
If you're reading AI policy debates online, assume some percentage of the "people" arguing are synthetic. The Turing test isn't an academic benchmark anymore. It's the price of entry for state actors running perception management at scale. Trust nothing. Verify sources. Follow the weird patterns.