TikTok removes AI images of Singaporeans missing in Nepal floods for spreading misinformation
SINGAPORE - A widely seen post on TikTok with AI images of Singaporeans missing in the Nepal floods has been removed after it violated the platform’s guidelines.
The post uploaded on Aug 31 contained AI-generated photos of four of the nine missing Singaporeans, alongside real photos of three of the other missing Singaporeans.
The Straits Times ran a story on the post on Sept 1, and TikTok removed it the next day.
The post had violated guidelines on edited media and generated AI content, spreading misinformation and deceiving users into thinking those were real images of the missing Singaporeans.
The post had over 150,000 views and close to 300 comments before it was taken down.
The missing Singaporeans have been uncontactable since Aug 26, when flash floods swept through parts of Tibet and Nepal.
The death toll as of Sept 3 was over 1,200, and over 5,000 are still missing.
The missing Singaporeans have been identified as Rengarajan Prabu, Shalinny Deavy Elan Sozan, Subramaniam Diraviam, Rajesvaran Ayyavoo, Estee Loeh, Ghosh Aditi, Ng Kit Kuen, Shubakshi Rawla and Sumit Rawla.
The TikTok post included real photos of Loeh and the Rawlas, and AI images of Rengarajan, Shalinny, Subramaniam and Rajesvaran.
Rengarajan, Shalinny, Subramaniam and Rajesvaran were travelling with an agency, Leaf Holidays Treks and Expedition.
On Aug 28, the travel agency posted photos of the four Singaporeans on Facebook.
It is believed the AI images of them on TikTok were derived from the travel agency’s photos.
In the fake images, Shalinny, who works at the State Courts, was portrayed with a black “pottu” (a traditional coloured dot or mark worn on the forehead).
Her husband, who only wanted to be known as Selvaganesh, had told ST he was upset when he saw the post, as the AI image did not look like his wife, who has only worn a red pottu, never black.
On Sept 4, Selvaganesh said he was relieved and thankful the post had been removed.
TikTok said it launches additional detection efforts with a dedicated task force in events such as natural disasters and public safety emergencies.
This led to it removing similar posts on the Nepal floods, alongside the one removed on Sept 2.
AI images on social media platforms are a growing problem, with more fake, AI-generated content entering Singapore’s online space.
AI has been used to impersonate senior political office holders here to scam victims, with the police warning the public to be wary of such ruses.
Detecting fake images
To fight the scourge of fake videos and images in Singapore’s online space, the Home Team Science and Technology Agency (HTX) has been developing tools such as AlchemiX to detect them.
HTX engineers ran the AI images of the missing Singaporeans through AlchemiX which detected traces of AI generation and face morphing.
Alvin Wong, from HTX’s Sensors and Sensemaking Centre of Expertise, said it may take just a few seconds to generate AI images, but these can spread very quickly and cause widespread harm in major events.
He said: “A polished AI-generated image may require several iterative prompts to perfect.
“But even one with obvious defects can spread rapidly during breaking news events, particularly when viewers are consuming content on small mobile screens where such flaws are harder to detect.”
He added that AI cannot accurately recover detail that is missing.
That means when one takes a blurry image and asks an AI to make it clearer and larger, the added details are not real.
He said: “The AI restoration is done on a purely predictive basis. Furthermore, the result of the restoration process can vary depending on the instructions given.”
Wong said with AI advancing rapidly, it is not possible for AI images to always be detected by tools.
But the public should take steps to check on the authenticity of a photo or video before deciding whether to share or report it.
He said things to check are:
Source: Check who first published the image. Look for the original post and any mentions of the photographer or organisation.
Context: Check whether the date, place, people and clothing match reliable records.
Image: Check whether the text, hands, shadows and object edges look consistent.
Other evidence: Check whether photographs or videos from trusted sources or official notices support the claim.
Wong said: “Find out who first published the image and what the post claims. Then, compare that claim with reliable records.”
KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.