- Actor
- PRC-linked content farms and coordinated inauthentic networks (Spamouflage / Dragonbridge lineage; commercial farms; Meta-attributed China-origin clusters). State direction vs state tolerance is high for the pattern, not always documented for a given page.
- Target
- Taiwanese audiences and the 2026 local/municipal election information environment; secondary Traditional Chinese diaspora
- Source
- Unidentified gray to black (lifestyle camouflage, proxy IPs, residual prompts that operators try to delete)
- Techniques
- Synthetic mediaNarrative launderingAstroturfingCard stackingFirehosingCorpus poisoning
- First seen (this wave’s forensic tell)
- Feb 2026 residual-prompt recovery (Austin H. Wang et al.); Meta H1 2026 Taiwan-proxy network; Sep 2026 amplification
- Status
- active as of September 10, 2026
- Related briefs
- China-linked content farms leave residual LLM prompts instructing AI to rewrite material specifically for Taiwanese audiences in Traditional Chinese, PRC-linked content farms and networks continue AI-assisted production of Taiwan-targeted Traditional Chinese narratives, with residual prompt leakage and proxy distribution patterns
Mechanism
The production line is: take a text, instruct a model to face Taiwanese users, rewrite in Traditional Chinese, keep a preferred historical frame, cap length, post through a farm or a lifestyle page, delete the prompt.
The load-bearing forensic object is a residual prompt recovered from a content-farm post and logged in the OECD AI Incidents Monitor: instructions to target Taiwan, use Traditional Chinese, hold “historical authenticity,” and stay near 500 words. The line was deleted within minutes and survived in edit history. That is an operator who knows the instruction must not be seen.
Meta’s first-half 2026 adversarial threat report separately described a China-origin network using Taiwan-based proxy IPs, mixed-currency ads (HKD/CNY/TWD), and pages that promoted pro-Beijing frames while criticizing the DPP. Earlier ASPI/Google Spamouflage work used AI news-anchor video in the 2024 cycle. Taipei Times (10 Sep 2026) noted comment floods with Mainland lexical markers (dayin, xinxi, zhiliang) consistent with Mainland-trained models.
27 March 2022. Entrance to the Meta Platforms headquarters complex in Menlo Park, California. Meta publishes the adversarial threat reports cited in this dossier. Photo: LPS.1, CC0.
Astroturf is the false local voice. Card-stacking is the preserved historical frame. Firehosing is the volume. Corpus poisoning is secondary: farm pages still get indexed and, in some cases, cited.
Results of the 2022 Republic of China county and city magistrate and mayoral elections, the previous local-election cycle before the 2026 local elections that this dossier's target environment concerns. The legend counts 14 Chinese Nationalist (Kuomintang), 5 Democratic Progressive, 1 Taiwan People's Party and 2 independent winners. Map: 沁水湾, Wikimedia Commons, CC BY-SA 4.0 (rasterised and flattened onto white by FRAME).
Evidence
- TechPolicy.Press, 8 Sep 2026, “China’s Disinformation Operations in Taiwan Are Changing with AI” — https://techpolicy.press/chinas-disinformation-operations-in-taiwan-are-changing-with-ai
- Austin H. Wang / Voicettank, 10 Feb 2026, residual prompts — https://voicettank.org/20260210-1/
- OECD AI Incidents Monitor — https://oecd.ai/en/incidents/2026-02-11-ea81
- Meta First Half 2026 Adversarial Threat Report — https://transparency.meta.com/sr/first-half-2026-Adversarial-threat-report/
- Taipei Times, 10 Sep 2026 — https://www.taipeitimes.com/News/editorials/archives/2026/09/10/2003863987
What would falsify this read
A showing that residual prompts are a farmer’s private workflow with no political targeting; that proxy-IP clusters are commercial SEO unrelated to cross-strait issues; or that Traditional Chinese lexical tells are just bilingual writers. “Every Taiwan comment flood is the CCP” is not a valid inference.
Literacy counter
Lifestyle-to-politics pivots, mixed simplified/traditional leakage, Mainland lexical tells, and identical comment storms around a local controversy are provisional. Cross-check against primary Taiwanese reporting. Platforms should retain edit-history and prompt artifacts as first-class indicators rather than relying on after-the-fact takedowns.
5 November 2021. Legislative Yuan Speaker You Si-kun and legislators receive a 13-member delegation of the European Parliament's Special Committee on Foreign Interference in all Democratic Processes in the EU, including Disinformation (INGE), led by MEP Raphaël Glucksmann, before the Legislative Yuan assembly hall in Taipei, per the Legislative Yuan's description. Photo: Legislative Yuan (立法院), Taiwan Government Website Open Information Announcement.