OpenAI Disrupts PRC-Linked 'Tech and Tariffs' Influence Operation

OpenAI has disrupted a coordinated influence operation, dubbed the ‘Tech and Tariffs’ campaign, involving a cluster of ChatGPT accounts likely originating in China. These accounts were used to generate political content, criticize US tech policies, and attempt to damage OpenAI’s reputation through false claims of data breaches.

Actor Identification and Attribution

OpenAI identified and banned a cluster of ChatGPT accounts that exhibited behavioral indicators consistent with actors aligned with the interests of the Chinese Communist Party (CCP). While a precise institutional affiliation could not be established, the attribution is based on several key factors:

  • Language and Access: The accounts prompted ChatGPT in Simplified Chinese and utilized VPNs to access the platform.
  • Terminology: Prompts frequently used language associated with China’s public security system, specifically seeking risk assessments for protests, police-related incidents, petitioning activity, and crowd movements in Shanghai.
  • Self-Identification: One user explicitly referred to their operated social media accounts as a ’water army’ (映关), a term for coordinated online accounts used to flood platforms with trolls or criticisms.
  • Explicit Intent: Users instructed ChatGPT to generate content specifically designed to benefit the PRC or advance pro-PRC narratives.

Campaign Behaviors and Content Generation

The operation utilized ChatGPT to produce a wide array of content for distribution via likely inauthentic X accounts, focusing on several strategic themes:

US-China Technological Competition

Actors framed the competition around AI, 5G, rare earths, new energy, and industrial resilience. They claimed the US was pursuing technological dominance and rule-making power. A specific directive was given to generate cartoons depicting President Trump, while explicitly forbidding any imagery of China or Chinese leader Xi Jinping.

Pro-PRC and Anti-Western Narratives

Beyond tech policy, the actors generated bulk quantities of short, colloquial comments in Chinese and English. These posts were designed to favor the PRC, attack the US and Israel, and amplify anti-Jewish tropes, including claims that ’Jewish capital manipulates public opinion.’ They also used the tool to harass Chinese dissidents.

Surveillance System Design

The actors requested a concept for an AI system to surveil online public opinion. This proposed system was intended to automatically scrape ’harmful’ information from ’key persons’ on social media, store logs, and perform large-scale semantic analysis of downloaded videos to send risk notifications. OpenAI reports that the models generated a general 500-word output on data storage and management but did not provide specific methods for collecting surveillance data.

Platform Operations and Coordination

The campaign targeted international audiences, producing content in English, Italian, Japanese, and Traditional Chinese (specifically targeting Taiwanese audiences). The operation was characterized by two interconnected clusters of inauthentic X accounts:

  1. The ’Tech and Tariffs’ Cluster: Focused on generating and posting the aforementioned political content.
  2. The ’ChatGPT Compromise’ Cluster: A network of accounts created in late 2025 that posted false claims that ChatGPT user data had been compromised.

OpenAI found that these two clusters interacted and amplified each other. For example, a post about US intelligence agencies allegedly hacking mobile networks generated by the ’Tech and Tariffs’ cluster was reposted by the ’ChatGPT compromise’ cluster. Both clusters also coordinated their activity by quote-tweeting the same unrelated verified account within hours of each other.

Impact and Strategic Context

Using the Breakout Scale, OpenAI assesses this activity as Category One, meaning the activity spanned one platform (X) with no evidence of breakout to other platforms. Most posts generated little to no observable engagement, and the false claims regarding data compromise were not amplified by high-reach authentic accounts.

Parallel to Rare Earths Campaigns

OpenAI notes that this pattern of targeting a private company in a strategic industry mirrors previous PRC-origin operations identified by Mandiant and the Australian Strategic Policy Institute (ASPI). Specifically, the Spamouflage/DRAGONBRIDGE network previously targeted companies like Lynas Rare Earths, Appia Rare Earths & Uranium, and USA Rare Earth to protect the PRC’s market dominance in strategic minerals.

Geopolitical Timing

The campaign coincided with a sharp escalation in US-China economic tension, specifically the announcement of an additional 100 percent tariff on Chinese goods by President Trump. It also followed the CCP’s Fourth Plenary Session, where the 15th Five-Year Plan elevated AI as a strategic technology and industrial priority through a nationwide ’AI+’ initiative.

Sources