On 12 October 2026 the Taipei Times reported that an official familiar with China's united-front tactics expects Beijing to run a short, intense disinformation burst, using deepfakes and bot accounts, to sway next month's Taipei mayoral election. Authorities said they would act "in accordance with the law" against anyone China mobilizes. The report named no platform measures. The only enforcement it cited was a prosecution and an association's dissolution, both aimed at people, not at content.
That gap is the story. Taiwan has strong tools against human agents of influence. Its tools against synthetic content on platforms are thin, and the main one has gone unused.
The strongest case for tougher action
The argument for a hard line is serious. Taiwan is the most intensively targeted democracy in the world for Chinese influence operations. Radio Free Asia reported on 31 August 2026 that the ruling DPP expects Beijing to use AI to impersonate Taiwanese social media users and circulate fabricated recordings or scandals shortly before voting on 28 November. Doublethink Lab found generative AI was used in the 2024 election for large-scale spam and to help inauthentic accounts vary their posts and schedules. When a fake lands days before a vote, "more speech" may arrive too late. A state that waits for perfect proof of foreign direction will lose the news cycle.
What enforcement has actually reached
The prosecution the Taipei Times cites is real. On 11 August 2026 the New Taipei District Court sentenced Xu Chunying to a combined seven years, according to Focus Taiwan. The sentence stacked several counts. Two were Anti-Infiltration Act violations for campaigning under the direction of Chinese officials, one for a mayoral candidate and one for a presidential candidate. The others were fraud-related and foreign-exchange offenses. The court found she acted on directives from named Chinese officials from 2021. Prosecutors had sought more than 20 years, the ruling can be appealed, and the Taiwan People's Party disputed the finding that Beijing directed her campaigning.
The association she led is the second data point. On 4 May 2026 the Taipei Times reported that the government planned to dissolve the Cross-Strait Marriage and Family Service Alliance under the Civil Associations Act, citing an unnamed source.
Both actions go after organizers and organizations, which is the right target for an operation that depends on human networks. The Anti-Infiltration Act is built around that. It reaches conduct done on the instruction, commission or funding of a foreign hostile force. Neither action touches a deepfake already circulating on a platform, and a deepfake needs no organizer to go viral.
The takedown mechanism nobody uses
Taiwan did legislate for synthetic content. Article 51-3 of the Public Officials Election and Recall Act, added in May 2023, lets a candidate who finds a deepfake of their voice or likeness request a police investigation. If police confirm it, the candidate can send the file to the broadcaster or platform, which must restrict or remove the content within two days of the request.
According to a 1 October 2026 analysis by Eddy Yen-Ting Lin in Tech Policy Press, the Central Election Commission confirmed zero requests as of May 2026. Lin identifies why:
- Narrow coverage: only candidates are protected, and only from the official election notice, typically 40 to 120 days before the vote.
- Police as arbiters: the deepfake determination rests with police, with no appeal, and lacks an objective standard.
- Jurisdictional friction: foreign platforms claim to be outside Taiwan's reach, so serving a takedown request in time is hard.
- Gaps in practice: a 24 July 2026 video using a voice nearly indistinguishable from President Lai Ching-te's fell outside the mechanism. Lai is not a candidate in these local elections, and the video fell outside the statutory window. It remains online.
The Ministry of Digital Affairs has launched a reporting site relaying police-verified requests to Meta, Google, LINE and TikTok. That route still depends on voluntary cooperation. The Central Election Commission is also setting up a task force and an online reporting platform, per RFA. Those are useful and unproven.
Why platform duties work elsewhere and not here
Taiwan's anti-fraud regime shows that platform obligations can bite when the statute gives the regulator a clear hook. Under the Fraud Crime Hazard Prevention Act, the Ministry of Digital Affairs fined Meta NT$15 million in July 2025, per Taiwan News, over its failure to be transparent about advertiser information in 23 cases reported by police. The reason is structural. The fraud law imposes verification duties on a defined class of large advertising platforms. The election deepfake clause instead depends on a case-by-case police finding for each piece of content.
A proportionate fix
The temptation after a warning like the Taipei Times report is to expand takedown powers. That would be a mistake. Giving police or a ministry faster, broader authority to declare election content false invites misuse against satire and opposition criticism, especially from whichever party holds office. Lin makes this point as well, noting that California's analogous deepfake laws were enjoined on First Amendment and Section 230 grounds.
A more defensible package would have four parts:
- Judicial review: make takedown decisions reviewable, with an appeal path, rather than final on a police finding.
- Wider time and subject coverage: protect against foreign-origin impersonation year-round, with narrow remedies that carve out satire.
- Domestic legal representatives: extend to major social platforms the representative requirement the fraud law already imposes on large advertising platforms, so that notices can be served.
- Transparency: publish how many reports are received, how many are verified and how fast platforms act. If the mechanism stays at zero uses, that is itself a finding.
Prosecuting Beijing's agents should continue under the existing statutes. But the Taipei mayoral race will test a different question: whether a democracy can respond in two days to a fake without giving officials a standing power to decide what is true. The evidence so far is that the response tools exist on paper and have not yet been exercised.