Operation Nine—emdash Line: Disrupting China-Origin Influence Activity

OpenAI has banned a network of ChatGPT accounts linked to a covert influence operation originating from China. The operation used AI to generate cross-platform social media content and conduct reconnaissance to influence political narratives in Southeast Asia, Hong Kong, and the United States.

Operation Overview and Target Narratives

The network focused on bulk-generating social media posts to support Chinese territorial claims in the South China Sea and West Philippines Sea, specifically referencing the "Nine-Dash Line" boundary. Because the operation frequently used em-dashes in its AI-generated text, OpenAI named the activity "Nine—emdash line."

Key target narratives included:

  • The Philippines: Generating posts criticizing President Ferdinand Marcos, including false allegations of drug use and election manipulation.
  • Vietnam: Creating content regarding Vietnam—specifically its alleged environmental impact in the South China Sea.
  • Hong Kong: Producing Cantonese-language posts to discredit pro-democracy activists and political figures such as Jimmy Lai, Nathan Law, and Agnes Chow, while expressing support for Hong Kong national security laws.
  • United States: Generating English-language content regarding political issues, including the fentanyl crisis.

AI Utilization for Content and Reconnaissance

The threat actors used ChatGPT for more than just text generation; they employed the models for research, strategic planning, and persona development.

Research and Reconnaissance

Operators used the models to identify niche blogs, forums, and less regulated social media platforms across Europe, America, and Southeast Asia. They also requested lists of common Tibetan names, a tactic OpenAI identifies as a method for creating inauthentic social media personas. OpenAI notes that this use of AI for research provided greater convenience but did not necessarily increase the actors' overall capabilities compared to traditional search engines.

Social Media Growth Strategies

In a novel application, the network sought advice on growth strategies, including how to initiate a TikTok challenge. Specifically, they attempted to leverage the existing #MyImmigrantStory hashtag by asking the model to ideate and generate transcripts for TikTok posts, as well as requesting recommendations for background music and imagery.

Technical Links and Attribution

While the social media accounts showed behavioral similarities to "Spamouflage"—a long-running China-origin operation—OpenAI did not identify direct technical links between the two. However, some videos used in the "Nine—emdash line" operation had been previously identified by the Australian Strategic Policy Institute as part of an operation sharing similarities with earlier Spamouflage activity.

Impact Assessment

OpenAI assesses the operation as Category 2 on the IO impact Breakout Scale, meaning it was active on multiple platforms but achieved minimal to no breakout engagement.

The operation's impact was limited by several factors:

  • Low Engagement: Most posts received minimal engagement, and many replies or reposts were generated by other accounts within the same controlled network.
  • Lack of Sophistication: The personas used stock images for profile photos or default handles and utilized coordinated behavioral traits common in other China-origin operations.
  • External Detection: Philstar.com independently discovered and reported on a subset of the network on X, which continued to operate even after the ChatGPT accounts were banned.

Sources