SINGAPORE: Since Singapore's May 3 polls were called, there has been a surge in the number of General Election-related online videos created using artificial intelligence (AI) tools.
CNA detected 73 such videos posted between Apr 15 and Apr 19. This was more than the 64 such clips in an entire month from Mar 14 to Apr 14. All the videos were posted on TikTok.
Apr 15 was when the Writ of Election was issued.
The upswing in such videos comes in spite of a new law banning digitally generated or manipulated content that misrepresents what candidates say or do, regardless of whether it is favourable or unfavourable to them.
GE2025 is the first time the legislation is in place.From Apr 15 to Nomination Day on Apr 23, the ban covers prospective candidates who have paid their election deposits and given the Elections Department consent to make their candidature public.
After nomination proceedings close, the ban will cover all candidates through the remainder of the election period.
Out of the 73 AI-generated, GE-related videos posted since Apr 15, CNA found that 11contain digitally generated or manipulated visuals of prospective candidates, potentially violating this new law.
The number grows to 27if the count includes videos that contain genuine footage of prospective candidates, but together with AI elements like avatars and text-to-speech voiceovers.
Under the law, corrective directions can be issued to individuals, requiring them to take down the content; or to social media providers asking them to disable access.
Failure to comply can lead to jail or a fine.
As of Monday (Apr 21), no corrective directions against deepfakes misrepresenting candidates have been issued.
The Ministry of Digital Development and Information (MDDI) told CNAon Apr 21 that the public will be notified if there are any issued.
The ministry said it was"closely monitoring Singapore's online space during the General Election".
"There are processes in place to monitor for and minimise the risk of election interference that can arise from the spread of prohibited content, such as deepfakes misrepresenting candidates," it said.
"There are dedicated teams standing up during the election period for this purpose, and they will work closely with social media services to act swiftly on prohibited content."
MDDI also said candidates and members of the public should remain "alert and discerning consumers of information".
The ministry did not answer CNA's questions on what and how many reports of AI-generated content it has received, investigations it is conducting and enforcement actions taken.
On Apr 23, two days after this article was published, MDDI told CNA that the government has looked into the AI-generated videos mentioned within.
Several of the TikTok accounts and videos covered in this article have since been taken down, and those that remained online were reviewed.
None of the videos are in breach of the election rules laid out in the Elections (Integrity of Online Advertising) (Amendment) Act, said the spokesperson.
The Returning Officer for the election also has not received any requests from prospective candidates in relation to these videos.
"In general, the new law does not apply to AI-generated or animated characters and cartoons, cosmetic alterations, or entertainment content such as memes and satire, which often contain unrealistic depictions of candidates," said an MDDI spokesperson.
The spokesperson also reminded candidates and parties to "campaign truthfully and responsibly".
WHAT'S OUT THERE
CNA monitors a variety of TikTok accounts including several that have been found to use generative AI videos, images or audio clips to create local political content.
The videos were discovered using a list of keywords associated with Singapore politics and the upcoming election.
Prime Minister Lawrence Wong features in seven out of the 11 potentially problematic videos.
The most widely circulated one is more than two minutes long, and features an AI-generated portrait of Mr Wong with a text-to-speech voiceover in Thai. It was posted by TikTok account @sparkupdate on Apr 15.
Captioned "Singapore's parliament dissolved, paving the way for general elections", this video has been played more than 70,000 times and received more than 3,700 reactions.
Other videos feature Senior Minister Lee Hsien Loong and Cabinet ministers Chee Hong Tat, Desmond Lee, Edwin Tong, Gan Kim Yong and Heng Swee Keat.
Opposition politicians such as Mr Pritam Singh and Ms Sylvia Lim of the Workers' Party and Mr Leong Mun Wai of the Progress Singapore Party appear in some as well.
Some videos do not conceal the fact that they are digitally manipulated. They depict the politicians in unlikely costumes such as football uniforms, imperial court robes or a Wonder Woman getup.
Only four of the clips were clearly labelled as AI-generated. Out of the remaining seven, four were posted by TikTok account @shining_shinshin.
One video by the account shows Mr Wong and Mr Singh appearing to embrace.
But in the majority of the 11 videos, the tone appears negative towards the Singapore government. Some of the clips are explicitly political commentary, with two by TikTok account @peopleschampionleague simulating debates between ruling and opposition politicians on issues such as the Goods and Services Tax (GST) hike and housing prices.
CNA earlier reported on a series of Chinese-language videos by TikTok account @ahboon87a, which can no longer be found on the platform.
These used Mr Lee Kuan Yew's death as an opening before launching into hot-button issues like immigration and transport, and questioning whether Singapore has achieved what its leaders promised.
The remainder of the 73 GE-related videos may not specifically contain AI-generated likenesses of politicians. But they still have the capacity to spread falsehoods.
For example, a Chinese-language video posted by TikTok account @aladingzk claimed that Singapore dissolved its parliament and called an election at 3am, describing it as a "political earthquake".
In fact, the president's proclamation to dissolve parliament was published at 3pm on Apr 15.
AI VIDEOS IN ELECTIONS
Communications researcher Saifuddin Ahmed said generative AI poses a significantly larger threat now compared to previous elections, as the barriers to creating realistic deepfakes and audio are lower.
"What once required days or weeks of technical work and specialised hardware can now be run on home computers or in the cloud behind anonymity," said the Nanyang Technological University assistant professor.
"No society is immune. A single, well-timed deepfake could still erode public confidence, and even swift fact-checking may not fully neutralise its effects among some. There is a need to be constantly vigilant."
Generative AI has already influenced elections in other countries.
In Pakistan, AI-generated videos that surfaced just before polling day last year falsely showed former Prime Minister Imran Khan urging supporters to boycott the vote.
Mr Khan also used AI to deliver a speech from prison, in what Asst Prof Saifuddin described as one of the first large-scale uses of generative AI in an election.
India's general election last year also saw a proliferation of deepfakes, and a similar flipside with Prime Minister Narendra Modi resharing an AI-generated clip of himself dancing onstage.
During Indonesia's presidential election, also last year, a deepfake of the late President Suharto endorsing the Golkar party went viral. His son-in-law Prabowo Subianto, who was Mr Suharto's son-in-law and had support from Golkar, eventually won the election.
Further away in the United States, a primary election in New Hampshire saw the use of deepfake robocalls mimicking former United States President Joe Biden telling his supporters not to vote, causing confusion.
WHAT TO LOOK OUT FOR
Associate Professor Natalie Pang of the National University of Singapore said voters should start by looking out for labels or information indicating something has been generated by AI.
Or they could educate themselves on what AI-generated content looks and sounds like.
"It is also important to think before sharing anything, especially during the election campaign," said the head of the university's communications and new media department.
Asst Prof Saifuddin said his research suggests Singaporeans are least likely to report deepfakes. "This needs to change, as community vigilance is equally important," he said.
Any sensational and viral content online should be treated with "healthy scepticism", and voters should verify who published it and when.
"In cases when they are unsure, running images and clips through reverse search tools or doing a quick fact check online is necessary," said Asst Prof Saifuddin, suggesting that voters rely on reputable news outlets orofficial channels of political parties.
"If something seems off in the video, including awkward lighting, mismatched lip movements or an unusual voice, they should flag it."
