On July 24, a video titled "Oil Cannot Defeat You" appeared online, produced by content creator "Principal Lai'er" — real name Wei Chun-yu, a former director of the KMT's Youth Department — at the request of Taipei Mayor Chiang Wan-an. It used an AI-cloned narration resembling President Lai Ching-te's voice to mock the government's handling of a tainted cooking-oil scandal. After public complaints, Taiwan's Criminal Investigation Bureau opened a probe, and its technical analysis concluded the narration bore "characteristics of AI-generated audio" (Taiwan News). The Interior Ministry said the video could implicate forgery, defamation, and the Personal Data Protection Act (PDPA) — three statutes written years before generative voice cloning existed.
A defensible instinct, applied through the wrong tools
Interior Minister Liu Shyh-fang's defense of the probe deserves a fair hearing. Taiwan sits under sustained information-warfare pressure from Beijing, and a synthetic recording of a sitting president commenting on a live domestic scandal is not ordinary parody — it is a national-security-adjacent risk if believed genuine. Liu's position, that "Taiwan protects freedom of speech" but impersonating another person's voice "is unacceptable," and that police should investigate regardless of which party is implicated, is a coherent statement of enforcement neutrality, not a partisan threat by itself.
The problem is the toolkit. Forgery statutes were written for physical documents and seals. Defamation requires proving reputational harm from a false statement of fact, which is a strained fit for a video the KMT insists was understood as satire. And Taiwan's own Personal Data Protection Act — whose Article 41 sets criminal penalties of up to five years' imprisonment and fines up to NT$1 million for unlawful use of personal data (PDPA, Art. 41, law.moj.gov.tw) — was built around data-broker misuse and identity theft, not a satirical voice clone in a political ad. Prosecutors are retrofitting 20th-century statutes onto a problem none of them anticipated, and the fit is visibly awkward.
The KMT's overreach argument isn't baseless — but it proves too much
Chiang Wan-an, who commissioned the video, called the Criminal Investigation Bureau's visit to Wei's home an act of state intimidation; the KMT caucus argued police moved faster against a meme than against the oil-contamination case the video was mocking. That timing complaint is fair and worth answering directly — selective enforcement speed against a critic of the government, real or perceived, is exactly the kind of appearance a democracy should avoid creating. But the KMT's broader claim — that AI impersonation of an elected official is simply protected political speech, full stop — proves too much. The DPP's response, that impersonating a president risks the public mistaking synthetic content for an official statement, is not a fringe concern; it is the same rationale democracies from Brussels to Seoul now write into deepfake-labeling law.
Taiwan has a template for getting this right — and a cautionary tale for getting it wrong
This is not Taiwan's first attempt at platform-side rules for synthetic and disinforming content. In 2022, the National Communications Commission drafted a Digital Intermediary Services Act explicitly targeting deepfakes and disinformation, only to withdraw it after backlash over vague scope and the compliance burden it would have imposed on small forums like PTT (New Bloom Magazine). That bill never returned. What Taiwan has instead is the AI Basic Act, a 20-article framework law that took effect January 14, 2026 and states a principle of "appropriate disclosure or labeling of AI-generated content" — but creates no penalty for skipping it and imposes no obligations on private platforms at all (Ministry of Digital Affairs).
That gap is precisely what an August 10 Taipei Times editorial identified, arguing Taiwan needs "new legal criteria" — labeling and provenance verification for AI content, digital watermarking, rapid platform reporting channels, and evidence-preservation rules — rather than either blanket bans or ad hoc prosecution under forgery and privacy law (Taipei Times). The editorial's central point is the right one: judging a deepfake case solely on how convincingly it mimics a real voice, as the current forgery-and-defamation approach effectively does, tells prosecutors nothing about whether the content was clearly labeled as satire, how it spread, or whether a platform had a reporting channel available before it went viral.
The proportionate path is legislation, not selective prosecution
The Wei Chun-yu case will likely resolve on its own facts — whether the video was adequately understood as satire, whether it caused demonstrable harm. But treating it as a template for future deepfake enforcement, via whichever pre-AI statute happens to apply, guarantees more Chiang Wan-an-style overreach complaints, because criminal investigation is discretionary and looks partisan whenever it targets the opposition. A narrow, AI-specific statute — mandatory labeling for synthetic media of public officials, a clear safe harbor for disclosed satire, statutory platform reporting duties — would let Taiwan police the genuine risk (mistaking a clone for the real president) without relitigating a president's honor claim through Personal Data Protection Act charges every time a campaign ad gets aggressive. Taiwan tried and shelved a broad platform law in 2022; it does not need to repeat that failure to get a much narrower deepfake-disclosure bill right in 2026.