On 7 October 2026, Communications Minister Fahmi Fadzil told Parliament in a written reply that the Malaysian Communications and Multimedia Commission (MCMC) helped remove 713 deepfakes and doctored images of political party leaders between 1 January 2025 and 15 September 2026. The question came from PAS Youth chief Afnan Hamimi Taib Azamudden (PN-Alor Setar), who asked whether enforcement is selective against people with certain political views. The Star's report says the ministry rejected that, saying MCMC acts on "legal provisions, elements of the offence, facts and evidence", whatever the political affiliation of those involved.
The strongest case for the ministry
Fabricated video or audio of a party leader is not ordinary political speech. A convincing fake released days before a vote can spread faster than any correction. Voters cannot easily check it themselves, and the person depicted never said the words. A government that did nothing would be failing the voters. The ministry also points to real process: complaints and platform community-guideline breaches trigger removals, prosecutions need the Public Prosecutor's written consent, and affected parties can go to the Appeal Tribunal and then seek judicial review in the High Court, according to The Star.
The numbers also give some reassurance on scale. Between January and June 2026, MCMC asked platforms to remove 13,122 deepfake-related posts, and about 12,353 (94%) came down, New Straits Times reported in July. Most of that volume is scams and impersonation. Political deepfakes, at 713 over roughly 20 months, are a small slice. On those figures, the claim that MCMC is mainly policing opposition critics is hard to sustain.
Why the reply does not settle the question
A reply that denies bias cannot show the absence of bias, and nothing in it lets anyone test the claim. As reported, it does not name the platforms involved or say how many requests were refused. It also gives no split by party, no breakdown of which removals rested on a legal provision and which on a platform's own rules, and no definition of "deepfake" as opposed to satire or an unflattering edit. The figure covers "party leaders", so it cannot show whether a deepfake of a government minister is treated like one of an opposition figure. Without that, 713 is a count, not evidence of even-handedness.
The legal base is also broad. Section 233 of the Communications and Multimedia Act 1998 has been through repeated litigation. The Court of Appeal ruled in August 2025 that "offensive" and "annoy" in Section 233(1)(a) were void. On 6 February 2026, in Government of Malaysia v Heidy Quah Gaik Li, the Federal Court partly reversed that, held the provision constitutional and read it narrowly around intent, according to a Lee Hishammuddin Allen & Gledhill analysis. The same analysis says the Federal Court found the prosecution of Heidy Quah unjustified and directed the Attorney-General to issue guidelines on the charging threshold. A provision that needed that kind of judicial narrowing is a thin basis for a takedown programme that will run through an election campaign.
The enforcement record shows the pattern. NST reports that MCMC investigated 574 false-content cases under Section 233 between 2022 and June 2026. Twenty-three were prosecuted. Twelve have concluded, with fines totalling RM79,000, and 31 cases were compounded for RM1.22 million. Compounding, which means paying a sum to avoid trial, is quick and cheap for the state. It also means most cases end without a court testing whether the content met the legal threshold.
The platform layer is where the speech risk sits
Of the reply's three stated grounds, two are not statutory: complaints and breaches of platforms' own community guidelines. MCMC's request is in practice a government flag that a platform's trust-and-safety team weighs against its own rules. The Online Safety Act 2025 (Act 866) has been in force since 1 January 2026, according to MCMC's own page for the Act, and it puts duties on licensed providers. A platform fearing licensing or regulatory consequences has an incentive to say yes to any ministry request. Over-removal is then invisible, because nobody sees the lawful post that quietly disappeared.
MCMC's stated posture also points toward more reach. In a 3 January 2026 media statement, the commission said misuse of AI to produce grossly offensive content is an offence under Section 233, and it urged all platforms accessible in Malaysia to put safeguards on AI-powered features. That statement concerned sexualised image manipulation, and nobody should confuse it with political regulation. But the same legal tool and the same channel are now being used for political deepfakes.
What proportionate looks like before GE16
GE16 is expected soon, and Malaysia has no election-specific deepfake rules. Four things would make the programme defensible without weakening it:
- Publish takedown data. Release regular figures on requests, removals, refusals and legal basis, split by party affiliation of the person depicted. If enforcement is even-handed, this is the cheapest way to prove it.
- Define the target narrowly. Limit election-period action to realistic synthetic media that falsely depicts a real person saying or doing something, and exclude satire, parody and commentary explicitly.
- Prefer labels to deletion. The Risk Mitigation Code, which NST says has required AI labelling since 1 June, gives a less speech-restrictive tool for content that is not plainly deceptive.
- Give speakers fast notice and appeal. The Appeal Tribunal exists, but a remedy that arrives after polling day is no remedy for a candidate or creator whose post was wrongly removed.
Deepfakes are a real threat to elections, and Malaysia is right to act. Publishing the numbers, defining the target and offering quick redress would let the government answer Afnan's question with evidence rather than assurance.