OpenAI has pulled the curtain back on a previously unreported Chinese influence operation, dubbing it “Nine—emdash line” after its target and a telltale AI quirk. The company banned a network of ChatGPT accounts it says generated a flood of cross-platform content meant to sway opinion on flashpoints from the South China Sea to Hong Kong and even US politics. The operation’s name is a winking nod to both China’s sweeping “Nine-Dash Line” territorial claim and the AI-generated em-dashes littering its posts.
The operation’s bread and butter was bulk-generating English-language social media posts criticizing Vietnam’s environmental impact in contested waters and slamming Philippine President Ferdinand Marcos with allegations of drug use and election manipulation. One video shared alongside the drug accusations was already denounced by the Philippine government as a deepfake. The actors didn’t stop at posts attacking rivals; they also sought to discredit Hong Kong pro-democracy figures like Jimmy Lai and Nathan Law by churning out Cantonese-language comments painting them as traitors. In one particularly cynical play, an operator posted a critical comment and then used the model to generate a reply praising the same activist.
Beyond pumping out propaganda, the operators used ChatGPT for reconnaissance, hunting down niche forums and blogs across Europe, America, and Southeast Asia. OpenAI notes this was for convenience, not a capability leap—it mirrored basic search engine use. A novel twist was the network’s hunt for virality: they asked the model for advice on launching TikTok challenges and generating scripts to hijack the long-standing #MyImmigrantStory hashtag. This shows an evolution from mere posting to a genuine attempt at engineering organic reach.
While the social media accounts bore a resemblance to the long-running “Spamouflage” operation, OpenAI found no direct technical links. Many X accounts distributing the content have already been suspended. The operation’s focus on leveraging a popular hashtag and experimenting with TikTok formats suggests threat actors are getting savvier about platform dynamics, even if their AI-generated content still trips over punctuation.
