OpenAI Dubs China-Linked Influence Op 'Uncle Spam' After AI-Generated Veteran Bots Flop

OpenAI has banned a cluster of ChatGPT accounts linked to a China-origin influence operation it’s calling “Uncle Spam,” following a tip from Meta. The operation didn’t push a single party line. Instead, it played both sides of divisive American issues—especially tariffs—flooding X and Bluesky with AI-written comments that were both supportive and critical. The goal, consistent with old-school “firehose of falsehood” tactics, wasn’t ideological conversion but deepening existing cracks in U.S. political discourse.

The network’s tradecraft was a messy mix of the sophisticated and the sloppy. The accounts queried OpenAI’s models for the best times to post to juice engagement, a clear signal of calculated amplification. They then used DALL-E to generate stylized logos as profile pictures for fictitious personas, zeroing in on a particularly charged archetype: U.S. veterans critical of the current administration. That choice suggests an attempt to weaponize a trusted identity for maximum political sting. Yet, it largely failed. The social media posts generated by these models received almost no authentic engagement, with most X posts getting zero likes or reposts.

Beyond content generation, the operators used ChatGPT for reconnaissance. They consistently requested code and methods to scrape personal data from social platforms, asking for help with tools like Tweepy, Nitter, and Bluesky’s public API. The goal was to pull user profiles and follower lists, then analyze that data to categorize individuals by personal characteristics and organizational ties. This shifts the activity from mere propaganda spam to a more targeted intelligence-gathering effort, even if the reach was minimal.

What’s striking is the sheer scale of the flop. While some fake accounts had follower counts in the thousands, those numbers didn’t translate into likes or shares—a disconnect that often points to bot-driven followers that provide a hollow appearance of influence. OpenAI placed the operation at Category 2 on its IO Impact Breakout Scale, meaning activity across multiple platforms but zero breakout into authentic conversation. It’s a reminder that access to generative AI doesn’t automatically buy you an audience, though the ambition to blend automated content with targeted data collection shows these operations are still refining their playbook, not abandoning it.