OpenAI announced the banning of ChatGPT accounts associated with a China-origin influence operation, which it named "Uncle Spam." This operation, detailed in OpenAI's June 2025 report, leveraged AI to create polarized social media content and conduct research on individuals and online communities within the U.S. political landscape. The discovery followed a lead from Meta.
The "Uncle Spam" network generated social media comments supporting both sides of divisive U.S. political topics, including tariffs, and used AI-generated profile images across platforms such as X and Bluesky. The accounts also sought optimal posting times to maximize audience engagement.
Key Points
- OpenAI banned ChatGPT accounts linked to a China-origin influence operation, "Uncle Spam," in June 2025.
- The operation used AI to generate polarized social media content and AI-generated profile images for platforms like X and Bluesky.
- Accounts requested optimal posting times to maximize audience engagement.
- The network used OpenAI's image generation capabilities to create stylized logos for fictitious personas, including "Veterans for Justice."
- The accounts consistently requested code and methods to extract personal data from social platforms using tools like Tweepy, Nitter, and Bluesky's public API.
- Models were used for guidance to analyze profile data for personal characteristics and organizational affiliations.
- The operation produced polarizing content across U.S. political discourse, often within the same operational sessions, to exploit existing divisions.
Context
According to OpenAI, the "Uncle Spam" operation's content volume and focus on American messaging led to its designation. This activity was assessed as Category 2 on the IO impact Breakout Scale, indicating activity on multiple platforms with minimal engagement. Most posts received few to no likes or reposts, despite some accounts having thousands of followers, which did not translate into post engagement.
OpenAI has also reported on other influence operations. In October 2024, it banned Iran-origin "STORM-2035" accounts that generated U.S. and U.K. election content. In October 2025, OpenAI banned accounts linked to a PRC-origin operation, "Nine—emdash Line," which created regional influence content about the South China Sea, Hong Kong, and U.S. politics. In May 2024, OpenAI banned Israel-origin "Zero Zeno" accounts that generated anti-Hamas, anti-Qatar, pro-Israel, anti-BJP, and pro-Histadrut content.
Why It Matters
This case highlights the ongoing use of AI models in influence operations to generate content, create personas, and conduct reconnaissance. It underscores the need for platform providers to detect and disrupt such activities, even when they achieve low engagement.
What To Do
- Note that influence operations may use AI for content generation, persona creation, and data reconnaissance.
- Watch for reports from platform providers like OpenAI on detected influence operations and their tactics.
- Consider that low engagement metrics do not always indicate a lack of intent or sophistication in influence campaigns.
- Review the June 2025 OpenAI report for further details on "Uncle Spam."