Chinese state actors deployed AI-generated deepfake videos targeting Taiwan's presidential election candidates to spread disinformation and influence voter behavior as part of a broader cognitive warfare campaign.
In the lead-up to Taiwan's January 13, 2024 presidential election, deepfake videos were created using artificial intelligence and circulated on Taiwanese social media platforms. One notable deepfake showed presidential candidate Lai Ching-te appearing to endorse his two rivals, with stilted speech patterns that eventually led to its identification as AI-generated content. Another deepfake video purported to show current president Tsai Ing-wen advertising cryptocurrency investments, using voice synthesis and terminology common in China rather than Taiwan. These incidents were part of a broader Chinese influence operation that Taiwanese intelligence authorities reported had intensified as the election approached. According to intelligence briefings, senior Chinese officials including Wang Huning convened meetings to coordinate disinformation efforts, with the publicity department and a psychological warfare unit called 'Base 311' spearheading influence campaigns. Taiwan's fact-checking organizations, including MyGoPen and Cofacts, detected and debunked these deepfakes as part of their ongoing monitoring efforts. The deepfakes were designed to manipulate public opinion and undermine trust in democratic processes, targeting Taiwan's population of 23.5 million people who were eligible to vote in the election.
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.
AI system
Due to a decision or action made by an AI system
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed