Taiwanese officials have flagged nearly 2,700 social-media accounts to Meta that they believe may be part of a Chinese-linked influence campaign ahead of local elections scheduled for November 28. Reuters reported on November 20 that the accounts were identified on Threads, Meta’s text-based social platform, and that officials considered them possible vehicles for coordinated activity. The accounts were reportedly created in batches, with most opened during the previous four months. The information was based on three government officials and internal reports reviewed by Reuters, rather than a public technical assessment released by Meta.
The suspected campaign was not described as a conventional effort to promote a single candidate. Taiwanese officials told Reuters that the activity appeared designed to deepen divisions, undermine confidence in democratic institutions and discourage people from voting. The officials said accounts could initially present themselves as ordinary local users, discuss everyday interests and later shift toward criticism of the government or candidates. That approach attempts to build credibility before introducing political messages, making coordinated activity harder to identify from an individual post alone.
Officials say the account clusters were created in batches
One government report cited by Reuters said that of roughly 2,400 Threads accounts flagged by Taiwan, 74% listed China as their location, about 25% did not disclose a location and around 1% listed Russia. The report grouped the accounts into 27 clusters with similar Chinese names and different numerical identifiers, which officials said suggested batch creation. These details are part of Taiwan’s assessment; they do not independently establish the identity of every account operator or prove that each account was controlled by the same organisation.
Meta said it had removed the accounts flagged by Taiwan, many of which had already been disabled by its own systems. A company spokesperson described them as low-quality fake accounts and said they had not posted about election-related topics and most had no followers. Meta said it would continue taking action against accounts that violated its policies. The difference between the government’s assessment and Meta’s description is important: the two sides agree that accounts were removed, but the public statements do not provide a shared conclusion about their reach or the extent to which they had influenced users.
Taiwan also told Meta that about 260 accounts had disseminated more than 2,150 AI-generated videos that officials considered part of the campaign. Reuters reported that the videos attracted approximately 2.35 million interactions, including comments and shares, according to an internal report. Many targeted the Democratic Progressive Party and its policies. One example involved videos about Taipei mayoral candidate Puma Shen that accused him of bullying when he was a student; Shen denied the allegations. The report does not establish that every interaction represented a genuine voter or that the content changed voting intentions.
AI-generated videos added scale to the alleged operation
The use of AI-generated video can make influence operations easier to scale because operators can produce variations of content without arranging conventional filming or editing for every clip. Synthetic media can also be combined with ordinary text posts, reposts and coordinated commenting to create the appearance of broader public discussion. But identifying a video as AI-generated does not by itself establish who created it, who distributed it or whether it reached a persuadable audience. Investigators need to connect content, accounts, infrastructure and behaviour before attributing an operation.
The local elections are politically significant because they are a measure of support for President Lai Ching-te’s Democratic Progressive Party ahead of presidential elections in early 2028. China claims Taiwan as its territory and rejects the ruling party as separatist, while Taiwan’s government rejects Beijing’s sovereignty claim. Those tensions provide the broader political context for the suspected campaign, but they do not prove that every critical post or account originated from a foreign operation. Genuine political disagreement and coordinated inauthentic behaviour must be assessed separately.
Meta’s removal of flagged accounts shows that the platform responded to the government’s reports, but the episode also raises questions about transparency. The company’s statement that many accounts had already been disabled suggests that its own detection systems had identified at least part of the activity. It did not publish the full account list, the criteria used to remove them or an estimate of how many users saw the videos. Without those details, it is difficult to compare the government’s reach estimate with the platform’s internal assessment.
Meta and Taiwan describe the reach differently
The timing of the suspected activity is also relevant. Reuters reported that almost all of the flagged accounts had been opened in the previous four months, placing their creation ahead of the November 28 vote rather than showing a sudden wave on election day. The officials’ account describes a pattern in which apparently ordinary posts were used to build credibility before political criticism appeared. That pattern can be difficult to distinguish from genuine users who change their views over time. Investigators So rely on signals such as account creation clusters, repeated naming patterns, shared content and coordinated behaviour, rather than political opinion alone.
The case shows the difficulty of protecting elections on platforms where accounts can be created quickly and content can be amplified through ordinary engagement. Responses require cooperation between election authorities, platform operators and independent researchers, but they also need safeguards against removing legitimate speech. Taiwan’s officials say the accounts were intended to inflame divisions and discourage participation; Meta says the accounts had little visible reach and had not posted election-related material. The investigation and platform disclosures will determine how those assessments are reconciled.