Taiwan's Central Election Commission (CEC) is setting up a specialized task force and an online platform where candidates can report suspected deepfakes before the November 28 local elections. According to Radio Free Asia's August 31 report, the commission will assess reported material and refer confirmed cases to online platforms for action. The report cited no statutory basis and gave no launch date.
The design is sensible. It is also narrow, and that narrowness is the main reason to support it.
The strongest case for the new desk
The case for acting is strong. A fabricated recording released days before a vote can spread faster than any correction. Candidates have no practical way to prove a negative within that window. RFA quotes analyst Sze-Fung Lee expecting AI to be "more involved" in the November elections, through deepfakes for defamation, coordinated inauthentic behavior and microtargeting. Taiwan's National Security Bureau reportedly counted an average of 2.63 million Chinese cyberattacks per day on critical infrastructure in 2025. A state facing that pressure has good reason to build an intake channel for election-related fakes.
A July 2026 incident shows the problem is real. As the Taipei Times reported, a video released on July 24 allegedly used AI to mimic President William Lai's voice criticizing the government's handling of a cooking-oil scandal.
The legal basis is probably already on the books
The RFA report names no statute, but Taiwan's election law already contains a deepfake mechanism. Article 51-3 of the Public Officials Election and Recall Act lets a candidate who finds a deepfake of their own voice or likeness on television or the internet request an investigation. The right runs from the public notice of the election until the day before election day. The Act defines a deepfake as the use of digital composites or other technological methods to create a form that convincingly performs speech and actions that are not those of the actual person.
The platform duty follows a police finding. Within two days of receiving a police investigation report confirming a deepfake, platforms must stop broadcasting, restrict browsing, or remove the voice and likeness.
This is my reading, not something the CEC has said. The new desk looks like an administrative front door to that existing process, not a new power. If so, it deserves some credit. It uses a statute with a definition, a candidate-only standing rule, a police verification step and a deadline. It does not create a freewheeling "misinformation" authority.
Why the narrow scope is a feature
Taiwan's design limits itself to fabricated recordings of identifiable candidates. That is about the easiest category of political speech to adjudicate. Either a person said the words or they did not. The remedy targets the fake, not the argument.
The RFA report is candid about the limits. The system is weaker against coordinated networks that post political opinions or unverifiable claims, and against content in closed LINE communities, which are less visible to officials, researchers and fact-checkers. Those limits are what a free-speech-respecting rule should have.
A state that tried to reach opinions, contested claims or private group chats would need broad discretion over what counts as false. That discretion invites selective use against opposition candidates and critics. It would also push platforms toward over-removal, because they face deadlines and sanctions but little penalty for deleting lawful speech.
The fraud-law precedent shows what to avoid
The desk sits beside Taiwan's Fraud Crime Hazard Prevention Act, in force since July 31, 2024. The Ministry of Digital Affairs now supervises six platforms from four companies: Google, YouTube, Line, Facebook, Instagram and TikTok. Regulated operators must remove, restrict or stop broadcasting fraudulent ads within 24 hours of notification by authorities. They must also verify advertiser identity, keep fraud-prevention plans and publish transparency reports. Commentary on the regime reports fines of NT$500,000 to NT$10 million, up to NT$100 million for severe breaches, and joint liability with advertisers. It also notes disclosure of whether an ad uses deepfake technology or AI-generated images.
Commercial fraud is a bad template for political speech. A scam ad has a measurable harm and no speaker whose views deserve protection. A political deepfake sits close to satire, criticism and campaign rhetoric. Heavy penalties and short deadlines are defensible for the first. For the second they push platforms to remove first and review later. The election-law approach, with its police verification, is the better calibrated of the two, and officials should not import fraud-style fines into election content.
What to watch before November 28
The design is sound, but the CEC has not yet told the public how it will work. Four points will decide whether it stays proportionate:
- Publish the rules. The CEC should state its statutory basis, intake criteria and response times before launch, and give the platform a stable public address.
- Keep a verification step. Candidate allegations should not trigger takedowns without the investigation the statute contemplates. Candidates have an obvious incentive to label damaging but authentic recordings as fakes.
- Report afterward. The CEC should publish the number of reports, confirmed cases and referrals, plus any appeals or reversals, so the public can check for over-removal.
- Don't stretch the remit to LINE groups or opinion networks. Weakness there is not a loophole to close with surveillance or broader takedown powers. Media literacy, fast rebuttals from fact-checkers and transparency from platforms are better tools than a mandate to police private chats.
Taiwan is under real and sustained information pressure, and its answer so far is measured. It targets verifiable fabrications about named candidates, relies on an existing statute and leaves opinion alone. The measure of success on November 28 is not how much content disappears. It is whether voters could still trust that what they saw of their candidates was real, and whether the process stayed too transparent and narrow for anyone to turn against legitimate speech.