Nearly 14,000 removals, with tougher platform action due
Malaysia has secured the removal of 13,883 deepfake items, including 1,178 involving political leaders, following public complaints and takedown requests from the Malaysian Communications and Multimedia Commission (MCMC), Communications Minister Fahmi Fadzil told Parliament on November 17.
Contents
- Nearly 14,000 removals, with tougher platform action due
- What platforms must change
- Complaints climbed sharply before the latest tally
- AI misuse investigations brought fines and warnings
- Why political voices are easy to imitate
- Why removal speed matters during elections
- What people can check before sharing
- Key Points
The removals cover January 1, 2024, to October 31, 2026. They come as the government gives digital platforms until the end of 2026 to establish mechanisms that automatically address deepfakes and false material uploaded repeatedly, rather than leaving users to report the same content again and again.
Fahmi said platforms that fail to comply with the Risk Mitigation Code could face penalties ranging from RM1 million to RM10 million for each offence. The code, which took effect on June 1, requires licensed providers to identify, assess and address harmful content, clearly label certain material created or manipulated using artificial intelligence, and make reporting tools accessible.
The figures place political impersonation within a wider problem. Items involving political leaders accounted for about 8.5% of the 13,883 removals, or roughly one in 12. The remaining 12,705 items were outside that category, although Fahmi did not provide a detailed breakdown of their subjects.
For people using social media, the practical concern is simple: a familiar face or voice can now appear in a convincing recording of something that person never said or did. The government response combines removal requests, rules for platforms, criminal investigations and a public campaign urging people to check information before sharing it.
What platforms must change
Fahmi identified repeated complaints without effective platform action as a central challenge for the government and MCMC. Removing one upload leaves a continuing problem when the same false recording returns in another post or through another account.
The deadline at the end of 2026 is intended to move platforms toward automatic action against this repeated circulation. Fahmi announced it while answering a supplementary question from Afnan Hamimi Taib Azamudden, the Perikatan Nasional MP for Alor Setar, about screening deepfakes before publication and identifying their creators.
The Risk Mitigation Code operates under the Online Safety Act 2025. It requires licensed providers to have mechanisms for identifying harmful material, judging the risks it creates and taking action. Its labelling requirement covers images, audio and video generated or manipulated using AI that could be mistaken for authentic content.
Fahmi explained the labelling obligation during Question Time:
"The code also requires images, audio or videos generated or manipulated using AI that could be mistaken for authentic content to be clearly labelled or marked," he said.
A label tells a viewer that a recording has been generated or changed. Removal addresses content that warrants taking down. The code also requires accessible reporting tools, giving users a route to alert platforms when harmful material reaches them.
Election periods receive particular attention under the code: platforms must account for heightened risks during such periods in their assessments. Malaysia's 16th General Election must be held by February 2028, making preparations before the campaign a pressing part of the debate.
Complaints climbed sharply before the latest tally
Deputy Communications Minister Teo Nie Ching previously told the Dewan Rakyat that MCMC received 917 complaints about deepfake content in 2024, rising to 3,612 in 2025. By June 15, 2026, it had received another 7,967.
The 2025 figure was almost four times the 2024 total. The number received by June 15, 2026, was already more than twice the total for all of 2025. This rapid increase helps explain the pressure on platforms to handle complaints faster and prevent repeated uploads.
Complaints count reports submitted to MCMC, while the latest removal figure counts content items taken down by providers. A complaint, an upload and a criminal investigation are different units: one recording can circulate through many posts, while an investigation can examine conduct involving several pieces of material.
Fahmi had previously given a figure of 713 removed posts involving political leaders for January 1, 2025, to September 15, 2026. The latest total of 1,178 covers a longer period, beginning in January 2024 and ending in October 2026.
The main dates in the response are:
- January 1, 2024: The period covered by the latest removal total begins.
- June 1, 2026: The Risk Mitigation Code takes effect.
- June 15, 2026: Deepfake complaints for the year reach 7,967.
- October 31, 2026: The latest tally closes at 13,883 removed items.
- November 17, 2026: Fahmi outlines the figures and platform obligations in Parliament.
- End of 2026: Platforms are expected to have automatic action mechanisms ready.
- February 2028: The deadline by which the 16th General Election must be held.
AI misuse investigations brought fines and warnings
MCMC investigated 61 cases involving AI misuse under Section 233 of the Communications and Multimedia Act 1998. Fahmi's enforcement summary placed the period between January 1, 2025, and October 31, 2026; he was also quoted as giving January 1 to October 31, 2026, for the same total.
Ten cases went to court. Six had been decided, resulting in fines totalling RM19,500, while four remained before the courts. No dates for those remaining proceedings were given.
Seven other cases received compound offers totalling RM755,000, and 17 warning letters were issued. A compound is a financial settlement offered through an enforcement process, separate from a fine imposed by a court.
Fahmi's fuller explanation placed the remaining 27 cases in two categories: investigations still under way, or cases classified as requiring no further action. He did not give the number in each category.
The minister said Section 233 can be used when the content meets the legal elements of an offence. The Online Safety Act and its code address platform responsibilities, while investigations under the Communications and Multimedia Act examine alleged offences. This gives the government routes to pursue both harmful material online and the conduct behind it.
Why political voices are easy to imitate
Siraj Jalil, president of the Malaysia Cyber Consumer Association (MCCA), warned that convincing political deepfakes can be made using a smartphone or basic laptop and a short recording of a candidate speaking.
He described how little material and equipment a creator may need:
"Anyone with a smartphone or basic laptop, a 15 to 30 second audio clip from a candidate's public speech and off-the-shelf generative AI tools can clone a voice and sync lip movements in minutes, for free or just a few ringgit," he said.
Voice cloning uses software to imitate how someone sounds. Lip synchronisation can then make a face appear to speak the invented words. A viewer may recognise both the face and voice, making the false statement feel familiar and believable.
Politicians supply abundant material through speeches, livestreams and press conferences, Siraj said. Those public recordings give people creating fakes clear examples of their voices, expressions and movements.
Some political clips are made for humour. Others make concrete false claims. During the Melaka state election campaign, Barisan Nasional deputy chairman Mohamad Hasan warned voters about an AI video on TikTok that falsely showed him promising RM25,000 to everyone. Such a clip can create expectations about money or policy using a promise the politician never made.
Why removal speed matters during elections
Siraj argued that platforms still rely too heavily on reacting to reports after videos have spread. A clip can remain online for hours or days while users flag it and moderators review it.
By then, people may have recorded their screens and forwarded copies into closed WhatsApp or Telegram groups. Taking down the original post leaves those copies circulating elsewhere.
The MCCA president described the urgency of responding during a campaign:
"During a compressed election campaign, speed is everything," he said.
Siraj called for platforms to use automated detection and visible watermarks when content is uploaded, rather than depending on users to declare that their posts were generated by AI. A watermark is a visible mark that helps viewers recognise synthetic material.
He also proposed dedicated election response teams and a deadline of under two hours to remove verified election deepfakes flagged by the Election Commission, candidates or accredited fact checkers. He called for a ban on paid promotion and recommendation systems boosting political content whose origin has not been verified.
Those measures are MCCA proposals. The government has set the deadline for automatic action mechanisms and described existing code obligations, while Siraj is pressing for more specific standards on speed, detection and distribution before the next general election.
What people can check before sharing
Siraj advised viewers to examine the mouth, teeth and jawline for slight blurring, unnatural blinking or teeth that appear unusually smooth during fast speech. Audio clues can include a flat or overly rhythmic voice, especially in local Malay accents, and sudden changes in background noise.
The account posting the material also deserves attention. Siraj said deepfakes often appear on newly created, anonymous or unverified pages rather than official candidate channels or established news outlets.
A practical check is to look for the same announcement on the named person's official channels and seek a fuller recording of the speech or event. If a clip promises money, announces a surprising policy or attributes an inflammatory statement to someone, verifying the claim before forwarding it can stop another copy from spreading.
MCMC's Safe Internet campaign is intended to build these habits. Fahmi said it covers deepfake detection, identity impersonation and false information generated using AI, alongside the broader task of improving digital literacy.
The communications minister described the campaign's purpose in Parliament:
"On the prevention front, the government, through the MCMC, is also implementing the Safe Internet campaign to raise public awareness and digital literacy, particularly in identifying deepfake content, identity impersonation, and AI-generated false information," he said during Question Time.
For users, the immediate steps are to pause, verify and report suspicious material through platform tools. For providers, the approaching deadline demands a response to repeated uploads as well as individual complaints. The next concrete test is whether those automatic mechanisms are ready by the end of 2026.
Key Points
- Platforms removed 13,883 deepfake items from January 2024 through October 2026 following MCMC requests based on public complaints.
- Political leaders were involved in 1,178 items, about 8.5% of the total.
- Licensed providers must label relevant AI content, assess harmful material and provide accessible reporting tools.
- Platforms have until the end of 2026 to establish automatic action mechanisms for deepfakes and repeatedly uploaded false material.
- MCMC investigated 61 AI misuse cases; six court decisions produced fines totalling RM19,500.
- Four court cases remained pending, while 27 cases were still under investigation or classified for no further action.
- MCCA is seeking faster election takedowns and stronger detection before the general election due by February 2028.






