A Secret History That Never Happened
Around Taiwan's January 2024 presidential vote, AI-generated videos pushed a fabricated “secret history” of the outgoing president, Tsai Ing-wen. Researchers who tracked the clips traced much of the material to Chinese software and to a network long tied to Beijing's online influence operations.
On 2 January 2024, days before Taiwan chose a new president, an anonymous user uploaded a lengthy document to Zenodo, an open-access research repository, titled “The Secret History of Tsai Ing-wen.” The file falsely accused the outgoing president of corruption and personal misconduct, according to the Australian Strategic Policy Institute, which analysed the campaign. Most of the document's metadata had been stripped, but enough remained to show it had been created with “WPS 文字,” a word processor made by the Chinese company Kingsoft and, ASPI wrote, “commonly used only in mainland China.”
What made the episode notable was what happened next. Between 4 and 10 January, ASPI reported, channels linked to a known influence network “posted at least 490 videos referencing the ‘secret history’ document” on YouTube before the platform suspended the accounts, according to its analysis by researcher Albert Zhang. Many featured AI-generated avatars presented as news anchors reciting the document's claims. At least one clip, ASPI found, used a “speaking portrait,” a photorealistic avatar generated by the U.S. company D-ID, and some videos had been edited with CapCut, software owned by the Chinese firm ByteDance.
Independent analysts documented the surge in real time. Austin Wang, an associate professor of political science at the University of Nevada, said more than 100 nearly identical videos had appeared on YouTube in a matter of days, reciting the same text and images. “There is no doubt this activity is coordinated and a show of force,” Wang said, according to the Taipei Times. The videos, the paper reported, were being reshared roughly 100 times a minute and were swiftly replaced when platforms removed them.
A network Beijing has used before
ASPI attributed the effort to accounts “likely linked to the CCP's largest network of inauthentic social media accounts,” a cluster researchers call Spamouflage or Dragonbridge. The network has been tracked for years across platforms, but the January videos marked an escalation in its use of generative tools to mass-produce content that imitated legitimate news.
The apparent aim, researchers said, was less to swing votes toward any single candidate than to erode confidence in the outgoing administration and in the vote itself. The volume of the clips, and their uniform scripts, pointed to automated production at scale rather than organic sharing.
The first documented case of its kind
Months later, Microsoft placed the Taiwan videos in a broader pattern. In an April 2024 assessment of Chinese influence operations, the company's Threat Analysis Center wrote that a Beijing-affiliated actor it tracks as Storm-1376 had deployed AI-generated content around the election, including fabricated audio and manipulated video of Taiwanese candidates. “This was the first time that Microsoft Threat Intelligence has witnessed a nation-state actor using AI content in attempts to influence a foreign election,” the company wrote.
The report described suspected AI-generated audio of the Foxconn founder Terry Gou — who had withdrawn from the race in November 2023 — appearing to endorse another candidate, an endorsement Microsoft said he had never made. The company said Storm-1376 had also promoted AI-generated memes attacking the eventual winner, William Lai, and had deployed AI-generated news anchors since at least February 2023.
Reach, but limited traction
For all the volume, researchers who tracked the operation said its measurable impact on the election was modest. Many of the fake-history videos drew only single-digit view counts before removal, the Taipei Times reported, and platforms took down channels as they surfaced. Discussion of the “secret history” faded in the days before the vote as officials and fact-checkers pushed back.
What alarmed analysts was less the outcome than the method: a template for flooding a democracy's information space with cheap, synthetic material timed to an election. As ASPI and Microsoft both noted, the tools used in Taiwan — off-the-shelf editors, avatar generators and mass-posting accounts — are widely available, and the tactics documented around the January vote are readily portable to other contests.