This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
In December 2025, a manipulated image of former Taiwanese legislator Kao Chia-yu was circulated online, replacing the Republic of China flag in a photo with the PRC five-star flag. The politician debunked the image, identifying it as AI-assisted disinformation intended to smear her political reputation ahead of the 2026 local elections. The incident highlights the growing prevalence of AI-generated or manipulated political disinformation in Taiwan.
In late December 2025, an altered image reportedly circulated online falsely depicting former Taiwanese legislator Kao Chia-yu posing in front of the PRC five-star flag, implying travel to or alignment with China. Reporting indicates the image was manipulated from a real photo taken at Taiwan's representative office in New York, replacing the Republic of China flag. Kao publicly debunked the image, citing AI-assisted disinformation and warning of political smearing ahead of elections.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Also classified as
The AI-manipulated image targeted a specific political figure, Kao Chia-yu, to damage her personal and professional reputation.
The incident involves the creation and dissemination of AI-manipulated political disinformation targeting Taiwanese politicians to influence public opinion and manipulate the upcoming 2026 local elections.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by human actors intentionally using deployed image manipulation tools to create political disinformation targeting a legislator.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
Taiwanese politicians were targeted with AI-assisted manipulated media, including a photoshopped flag and a fake video, to damage their reputation ahead of local elections. The low sophistication of the edits led to quick detection, resulting in a minor national security impact, though it highlights the ongoing threat of digital disinformation targeting democratic processes.
The AI system (image manipulation/deepfake software) was used as a tool by human actors to perform a specific function (altering a photo and generating a fake video) based on direct human inputs.
Limited Risk: The report describes AI-generated content such as deepfakes ('合成影像' or synthetic images/videos), which fall under Limited Risk and require transparency obligations to notify users of AI generation.
The primary incident involved manipulating a photo of Kao Chia-yu (Deepfake Image Generation), and the report also mentions a fake video of Wang Shih-chien (Deepfake Video Generation).
The report clearly details the nature of the disinformation, the specific politicians targeted, the exact manipulation performed on the image, and the political context (the 2026 elections). The facts are straightforward and corroborated by the politician's own social media response.
None. The incident is clearly a politically motivated disinformation attempt using manipulated media.
Technology
AI Video Forgery Tools
Malicious actors used AI to forge politician identities in videos.
Image Manipulation Software
Easy access to photo editing tools allowed altering background flags.
Data Inputs
Publicly Available Photos
Official photos of politicians were easily sourced and manipulated.
Identity Data Exploitation
Biometric and voice data of politicians used to train AI models.
Human Factors
Malicious Intent of Actors
Unscrupulous individuals deliberately spread political disinformation.
Credulous Public Auditing
Some citizens easily believe and spread unverified online content.
Process and Methods
Ineffective Verification
Lack of quick verification methods for public figures to debunk fakes.
Sloppy Editing Work
Manipulators left errors like the plum blossom emblem on the wall.
Management
Social Media Platform Control
Platforms failed to proactively detect and flag manipulated media.
Regulatory Environment
Weak Disinformation Laws
Insufficient legal deterrents for generating political deepfakes.
AI system involved
An AI system was involved, but the reports do not identify it.
Identified by automated extraction; can differ from the alleged developers above.
