Taiwan’s intelligence community is sounding an unusually loud and specific alarm as voters prepare to go to the polls in local elections on 28 November 2026. On 20 October 2026, the National Security Bureau sent a written report to the Legislative Yuan, ahead of a session of the Foreign Affairs and National Defense Committee, where Director-General Tsai Ming-yen is expected to face questions from lawmakers. The report, made public through Taiwan’s Central News Agency, says Beijing is pouring more effort, money, and machine power into shaping the island’s public conversation than ever before. Behind the dry bureaucratic phrase “disputed messages” lies a flood of online noise, fake identities, manipulated videos, and carefully targeted stories designed to confuse, divide, and steer voters. The bureau counted about 2.5 million such messages between January and September 2026, up from 1.69 million in the same period a year earlier—a jump of roughly 48 percent. That is not a small statistical blip; it is an accelerating wave. The messages did not all look the same. Some attacked government policy on energy, food safety, and defense, trying to make governance look chaotic. Others promoted particular candidates or praised local infrastructure, apparently hoping to shape how people vote in the “nine-in-one” elections. China, as always, rejects these claims and insists it has no interest in interfering in Taiwan’s elections or spreading disinformation. But Taiwan’s national security officials say the evidence is now too broad and too systematic to ignore. The real point, they stress, is not simply that unpleasant things are being posted online, but that the whole operation is becoming more industrial, more personalized, and harder to spot before it does its damage.
The channels used for this influence campaign are not obscure corners of the internet. Facebook remains the single biggest arena, and its share of the disputed messages jumped from 43.6 percent to 68.9 percent. Threads, the more text-focused platform, doubled from 4.5 percent to 9 percent. Online forums and YouTube also played a role. The bureau singled out Facebook and Threads as the main highways for what it calls China’s cognitive warfare—a term that sounds technical but describes something very ordinary: the effort to get inside people’s minds and reshape what they believe about their country, their neighbors, and their own choices. The campaign, according to the report, works like an assembly line. Official Chinese bodies or state media set the narrative frame, then hand it off to proxy accounts, online “water armies,” public relations firms, content farms, fake websites, and AI-generated audio and video. Fake accounts flood comment sections, boosting some voices and drowning out others. It is not meant to convince everyone; it only needs to create enough confusion that people stop trusting anything. The report also says China is actively gathering information on current events, candidates’ policies, campaign trends, and polling data, using both technical means and human sources. That intelligence is then fed back into the propaganda machine so the next round of messages can be more effective. In other words, Beijing is not just broadcasting generic political attacks; it is listening, learning, and adapting its strategy based on what seems to work. The goal is to influence voters’ preferences and, ultimately, the outcome of the election, by making the political environment feel toxic, unreliable, and exhausting.
The election, however, is only part of the picture. The NSB also reported a sharp increase in cyber intrusions aimed at Taiwan’s Government Service Network, the digital backbone of public administration. Around 2.58 million intrusion attempts were detected each day on average, with the pace rising since June. In July alone, total detections exceeded 100 million. The bureau said it could not rule out that these attempts were part of advance intelligence-gathering, preparation for later attacks, or direct efforts to interfere with the election. Cybersecurity in this context is not just about protecting databases; it is about protecting the integrity of democratic procedures. At the same time, China has stepped up what analysts call grey-zone activity in the seas around Taiwan. From June onward, coast guard and maritime patrols were extended into waters east of Taiwan and the western Pacific. Beijing justified this by pointing to maritime delimitation talks between Japan and the Philippines, but Taiwan’s intelligence officials read it differently: a slow, steady push to create an impression of Chinese “maritime jurisdiction” over the area. These patrols were accompanied by disputed messaging meant to normalize China’s presence. On land, the report says, there were ongoing attempts to infiltrate Taiwan’s military units and government agencies through intelligence contacts and online recruitment, all aimed at obtaining sensitive information. The overall picture is not one of a single dramatic cyberattack or a sudden fake-news storm, but of a relentless, layered operation running every day, using every available tool, and constantly probing for weaknesses.
Behind the numbers, Taiwan’s national security sources tell a more human story. In a separate report on 20 October, CNA reporter Wu Shu-wei quoted sources who described this year as one of “AI united front” work. The phrase blends the old idea of a united front—winning hearts and minds through many small contacts—with the new power of artificial intelligence. According to the sources, China is using AI to mass-produce fake accounts, analyze audience segments with precision, and deliver customized messages to different groups of Taiwanese people. The aim is to inflame divisions, both between political camps and across society more broadly. These recent operations, the sources said, are consistent with leaked documents from the Chinese company GoLaxy, which had previously been exposed by foreign media. The documents reportedly showed that Taiwanese social media account details and household registration data were being collected systematically. AI systems then sorted Taiwanese people into four categories: hardliners, friendly, swing, and objective. Each category received its own tailored mix of propaganda and attacks. Hardliners might be pushed toward more extreme positions; friendly people might be gently reinforced; swing voters might be given reasons to stay home; objective types might be flooded with contradictory information to make objective judgment seem impossible. The sources gave a striking example of how quickly a fake platform can switch purpose: a page that usually discussed health and wellness would, at a chosen moment, suddenly begin spreading claims that Taiwan’s Ministry of Health and Welfare did not care about the public. The switch is deliberate. Operators build trust and organic audiences over time, then weaponize that trust when it matters most.
The sources also outlined four features that make this new wave of influence operations especially dangerous. First, tailored narratives are now the norm. Instead of pushing one clumsy story to everyone, operators use audience segmentation to speak to each person’s existing interests, fears, and political leanings. Second, there is personalization and infiltration of online echo chambers. Realistic fake accounts mirror the views, hobbies, and language of the people they want to influence. They spend weeks or months behaving like normal community members before posting extreme statements during specific events, hoping to spark infighting and drive wedges between natural allies. Third, the operations are designed to suppress political participation. Messages suggest that a certain candidate is hopeless, that voting is pointless, or that the whole system is rigged. This kind of quiet despair can be just as powerful as anger; it keeps supporters at home. The sources noted this tactic is applied across the political spectrum, not exclusively to one side. Fourth, there is the fabrication of livelihood issues, which are close to people’s daily worries. Examples include false claims that certain parties’ candidates would allow large quantities of controversial drugs into the country, or would relax restrictions on nightlife and adult entertainment businesses. These stories are designed to provoke alarm among parents and to push moderate citizens into angry reactions. The emotional impact, not the factual accuracy, is what matters.
Taken together, the reports paint an unsettling portrait of a cognitive battlefield where truth is the first casualty and artificial intelligence is the weapon of choice. The sources warn that many Threads accounts that appear to belong to ordinary Taiwanese users are actually registered with phone numbers from other countries. On the surface they look local, friendly, and reliable; underneath, they are part of a coordinated machine. What makes this difficult is that the old defenses—spotting bots, detecting coordinated behavior, identifying propaganda themes—no longer work well. AI-generated personas are harder to distinguish from real people. They write in natural language, follow real users, respond to comments, and build history. They can be created faster than they can be taken down, and they can adapt during a campaign, learning which stories gain traction and which ones fail. The national security sources said this combination of big-data analysis, AI-generated personas, and automated account cultivation is becoming increasingly sophisticated and requires continued monitoring and response. But the human dimension should not be lost: behind the talk of frames, water armies, and intrusion attempts, real people are being manipulated into distrusting one another, questioning their institutions, and losing faith in the democratic process. The point of a report like this is not to generate panic, but to encourage awareness. When voters know that a wave of synthetic emotion is being aimed at them, they can pause, ask questions, and refuse to be herded. The election may be local, but the challenge it reflects is global: in an age of AI, the battle for attention is also a battle for trust.