Busted by Bad Reviews! OpenAI Disrupts China-Origin Influence Operation "Sneer Review"

Hi, it's Shiichan! Today I've got a bit of an unsettling one from OpenAI's misuse report, so let's dig in.

OpenAI News openai.com

What was announced?

OpenAI's News reported that it banned a cluster of ChatGPT accounts tied to what looks like a covert influence operation. Most of the prompts were in Chinese and focused on political and geopolitical topics relevant to China. One user even claimed in a prompt to work for China's Propaganda Department, though OpenAI says it has no independent evidence to confirm that claim.

The operation's most distinctive move was generating dozens of critical comments in Chinese about a Taiwan-centric video game called "Reversed Front," followed by a long-form article falsely claiming it had sparked major backlash. That's what earned the operation its name: "Sneer Review."

Why it matters

Generative AI is great at writing, which makes it easy to misuse for everything from short social posts to documents that look like internal reports. What stands out here is how varied the tactics were: mocking a video game, harassing an individual activist, and weighing in on policy debates, all under one operation. The accounts also followed a consistent pattern of a "main" account posting first, followed by reply comments from other accounts, to fake the appearance of organic engagement. Whether AI providers can catch this kind of manufactured engagement matters a lot for how much we can trust what we see on social media.

What changes

With the accounts banned, this operation can no longer generate new content through ChatGPT. OpenAI assesses that it disrupted the operation while it was still in an early, low-reach stage: neither of its fake Facebook Pages had any followers, and some of its Reddit posts were removed or blocked by the platform's own filters.

Using the Breakout Scale, a framework for measuring the impact of influence operations, OpenAI rated this one at the low end of Category 3, assuming the engagement figures on X and TikTok were genuine. It also notes that since much of the commenting activity itself was generated by this same network, that rating would likely drop further if more of the likes and views turn out to be inauthentic too.

Dive Deep

The operation ran two main workstreams:

• Mass-producing short comments in English and Chinese (with a few in Urdu), posted mainly on TikTok and X, with some spillover to Reddit, Facebook, and various websites
• Generating what look like internal documents, including an essay written in the style of an official public security document and a performance review describing how the operation was set up and run

The commenting pattern was consistent: a "main" account, often created just for that one post, would post first, then other accounts would pile on with replies to simulate organic buzz. On TikTok, screen names were often mismatched with the language of the content: an account with a Korean name posted in Urdu, while accounts with Thai and Hebrew names posted in English. On X, the accounts tended to use cartoon profile pictures and sometimes crypto-flavored names. On Facebook, the network created two Pages posing as news outlets, but neither had any followers or likes.

The content clustered around three themes:

• Dozens of Chinese-language attack comments on "Reversed Front," a Taiwan-themed game imagining resistance against the Chinese Communist Party, plus a follow-up article falsely claiming it had drawn widespread backlash. This is what gave the operation its name, "Sneer Review"
• False, harassing comments in English and Urdu targeting Mahrang Baloch, a Pakistani activist who has publicly criticized China's investments in Balochistan, including a video with a fabricated claim about her
• Comments both praising and criticizing the closure of USAID (the U.S. Agency for International Development), posted on TikTok and X

On engagement, two TikTok videos together racked up about 25,000 likes, and tweets from the main account typically pulled in around 10,000 views each. But OpenAI cautions that since many of the comments on these posts were themselves generated by this network, those numbers likely include a fair amount of inauthentic engagement and should be treated with caution.

Wrap-up

• OpenAI banned ChatGPT accounts linked to a covert influence operation believed to originate from China
• The operation was dubbed "Sneer Review" after its campaign of fake bad reviews targeting the Taiwan-themed game "Reversed Front"
• It ran two workstreams: mass short-comment generation (mainly TikTok and X, with some Reddit, Facebook, and website activity) and drafting internal-looking documents
• Targets included the Taiwan game, harassment of Pakistani activist Mahrang Baloch, and comments both for and against the USAID closure
• Real-world reach looks limited so far; OpenAI rated it at the low end of Category 3 on the Breakout Scale and flagged that much of the engagement itself may be inauthentic

If you're into how AI gets misused for influence operations and how platforms coordinate to catch it, this report is worth a read!