AI-Generated Deepfake Videos Used in Scams and Misinformation in Taiwan
AI-generated deepfake videos are being used in Taiwan for investment fraud, fake medical endorsements, and misleading content through celebrity impersonation, virtual avatars, and stolen videos. Taiwan's cybersecurity agency identified four prevalent tactics and provided detection methods based on technical flaws in AI-generated content.
What happened
AI-generated deepfake videos are increasingly used in Taiwan for fraud and misinformation through four primary methods: celebrity deepfake pornography created using face-swapping technology, impersonation of financial experts or medical authorities to promote investments or medications, AI-generated virtual avatar content farms (such as a fake "Doctor Chen Zhiming" channel on YouTube), and reposting AI-generated videos from Sora with watermarks removed or cropped.
Reported impact
Scammers use AI deepfake videos impersonating financial and medical authorities to deceive people into joining investment groups or purchasing products. Virtual avatar content farms produce high volumes of misleading content on topics including health, finance, and elderly welfare. Celebrity deepfake pornography and misleading short videos proliferate on social media platforms.
Index analysis
Taiwan's National Institute of Cyber Security identified that AI deepfake videos contain detectable imperfections in facial features, eye movements, teeth and fingers, contours, facial expressions, voice-subtitle synchronization, lighting, shadows, text errors, spatial structures, camera movements, and object details. AI currently cannot perfectly generate straight lines. Virtual avatar videos often show periodic repetition of hand gestures despite appearing natural.
Governance controls
Multiple social media platforms including YouTube have introduced AI content disclosure labels requiring creators to mark AI-generated content. Users can verify content through three methods: checking for AI generation labels, reverse-image searching to confirm person identity and verify statements, and assessing whether the scenario context (people, events, time, location, objects) is reasonable.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.