AI-generated women are spreading disinformation about Singapore on TikTok

Singapore

A CNA investigation has found about 500 TikTok videos pushing false or misleading claims about Singapore or Malaysia, drawing a total of more than 3 million views.

New: You can now listen to articles.

This audio is generated by an AI tool.

13 Jul 2026 06:00PM

(Updated: 13 Jul 2026 06:32PM)

SINGAPORE: With their perfect hair, polished makeup and poised demeanour, the women could not be more ready for the bright lights of social media. Using TikTok as their stage, they speak confidently in Mandarin about weighty issues concerning Singapore and Malaysia, from the economy to geopolitics and foreign relations.

But many of these “presenters” are not what they seem.

A CNA investigation of 30 such TikTok accounts found a factory-like system behind over 550 videos. Nearly all (98 per cent) were assembled by taking AI-generated, manipulated or copied female personas, and stitching them together with reused voices and recycled scripts. Nearly nine in 10 videos pushed false or misleading claims about Singapore and Malaysia, drawing more than 3 million views in total.

Between October 2025 and June 2026, they repeatedly advanced a worldview where Singapore’s fortunes were dictated by China. The narratives blended fact, distortion and outright invention: The most outlandish one claimed that Singapore’s foreign minister Vivian Balakrishnan had unsuccessfully begged China and Indonesia not to let a new shipping route bypass Singapore’s port. No such thing happened.

As part of its investigation, CNA sent questions and shared examples of two accounts to TikTok. A few days later, both accounts were terminated.

A spokesperson for the platform said the accounts had broken rules on “deceptive behaviour”. She added that TikTok constantly monitors for and removes accounts trying to “manipulate our systems or our community in order to influence public debate”.

According to TikTok, the range of deceptive behaviour includes covert influence operations, impersonation, spam and fake reviews.

FACT AS BAIT

Unlike overtly hostile information campaigns, these TikTok videos at first glance portrayed a facade of normal, even harmless financial or geopolitical commentary.

This was in part due to the videos’ creators’ decision to use genuine news as hooks for baseless claims. One alleged that Singapore’s port dominance was about to “collapse” now that China’s Hainan Free Trade Port had become fully operational in December. Another seized on an unrelated event, ExxonMobil’s Singapore plant closure, as supposed proof of Singapore’s decline.

In fact, Singapore’s port handled a record of 44.66 million containers in 2025, and has ranked as the world’s second-busiest leading container port for the last 15 years. A previous CNA investigation in February found similar narratives being pushed on YouTube, suggesting such operations are not confined to one platform.

Experts told CNA they have observed similar ongoing campaigns on other social media platforms, pushing the same underlying content.

In CNA’s TikTok investigation, another set of videos used the same tactic – of trying to blur the line between truth and falsehood – by leaning on how Singapore has been China’s largest foreign investor since 2013. This real statistic served, however, as the foundation for subsequently arguing that Singapore had prospered by aligning itself with China’s rise, and for implying that other countries should do the same.

But Singapore is not the only target of the disinformation campaign.

Where the Singapore-focused claims cast China as a patron, the Malaysia ones took on a more contentious framing, with several blaming the country’s allegedly slower development on policies that sidelined its ethnic Chinese population.

The goal appears to be to erode trust and fracture social cohesion, said Associate Professor Saifuddin Ahmed from Nanyang Technological University (NTU)’s Wee Kim Wee School of Communication and Information.

“Any audience that grows to distrust its own government's narrative is more susceptible to external influence over time,” he added.

The experts also suggested that beyond domestic Singapore and Malaysia audiences, the global Mandarin-speaking diaspora could be a target. There is strategic value, for instance, in attempting to shape how investors might perceive the stability and political orientation of the two countries, said Assoc Prof Saifuddin.

INFRASTRUCTURE FOR EXPOSURE

Of the more than 550 videos CNA analysed, 94 videos were found to be consistently repeating false or misleading claims. This subset accounted for over 1.6 million views, almost half the total views in the investigation. In other words, the most heavily repeated scripts were also among the most watched.

These 94 videos appeared to adopt a similar playbook: Different female presenters “took turns” delivering the same commentary over weeks and months, creating an illusion of independent voices arriving at the same conclusions.

In all, 24 accounts parroted the same talking points at different times. The fabricated Dr Balakrishnan anecdote alone was repeated over nearly two months and viewed more than 100,000 times.

“TikTok's short-form, high-volume format means users encounter the same narratives repeatedly across different accounts without necessarily recognising it as a coordinated campaign,” Assoc Prof Saifuddin explained. “That repetition alone, independent of content quality, increases the likelihood that claims feel familiar and therefore plausible.”

Experts said the use of low-cost, attractive AI-generated presenters was also a known tactic to increase engagement, attention and trust.

“If viewers are more likely to watch, share and return to content fronted by an appealing presenter, that's a straightforward amplification strategy regardless of the underlying message,” said Assoc Prof Saifuddin.

“It exploits the same psychological mechanisms that make influencer content effective,” he added. “The female presenter choice may equally reflect what performs best on these platforms algorithmically.”

The political communication scholar also said it was believable that male viewers were being specifically targeted.

The same tactic has been observed in previously documented influence as well as marketing campaigns. According to reports in 2024, Chinese state-linked groups had targeted the Taiwanese elections that year with AI-generated news anchors. And in June this year, a network of fake female dating profiles were found pushing propaganda ahead of local polls in Taiwan.

CONVEYOR BELT DISINFO

Reusable scripts were just one facet of the machinery at work.

Among the accounts which shared scripts and talking points, half of them used the same audio tracks. Lip-sync errors were also common, and added to the possibility that the videos’ audio tracks had been produced separately before being spliced onto the visuals.

In many instances, identical-looking presenters were also found speaking in different voices across multiple videos. Some of these presenters appeared to be outright copies of other creators.

The biggest clue to their synthetic nature was how their heads or torsos were almost unmoving and pinned in place, within a single video and across multiple videos. In real footage, speakers naturally shift and adjust their posture as they talk.

Other signs pointed to possibly coordinated behaviour behind the accounts: Several followed a similar naming template of pairing a persona’s moniker with a finance-related term, while also demonstrating “burst posting” – or publishing multiple videos in a narrow window of time.

Taken together, the evidence leaves the accounts looking less like a group of unrelated, independent creators than a single production line.

CNA however could not identify an operator or paying client behind the operation, nor determine whether it was commercial or state-linked.

DON'T PLAY THE "LOSING GAME"

Asked about the effectiveness of the videos, Assoc Prof Saifuddin said that “view counts tell you about reach, not about whether minds were actually changed”.

But he also reiterated that the operation “doesn't need to convince people outright; it just needs to make certain framings feel familiar and recurring”.

The use of Mandarin is also a factor, with fact-checking and debunking efforts in this space lagging behind English-language cases, said Mr Benjamin Ang, who heads the centre of excellence for national security at the S Rajaratnam School of International Studies (RSIS) think-tank.

He urged people to highlight to friends and family the harms of being “drawn into the highly addictive habit of doomscrolling … so that we do not contribute to the toxic business model of these inauthentic content creators”.

Assoc Prof Saifuddin meanwhile described trying to spot AI artifacts as a “losing game”, given rapid advances in deepfake technology.

What's needed instead is greater awareness of the “fingerprints” of coordinated operations – such as repetitive claims not reported elsewhere – alongside more “prebunking” or prior explanation of these techniques before people encounter them, he added.

“Inoculating people against the method rather than specific claims is more scalable,” said Assoc Prof Saifuddin.

“Because the content changes, but the playbook doesn't.”

Source: CNA/jo