Recognising Deepfake Scam Videos That Impersonate Celebrities and Politicians

If you receive a video of Prime Minister Lawrence Wong, or Deputy Prime Minister Gan Kim Yong, or a celebrity you recognise from the hawker centres, telling you about a once-in-a-lifetime cryptocurrency opportunity, stop. Do not click. This is a deepfake. The Singapore Police Force Anti-Scam Command reported that in the first half of 2024 alone, there were 13 cases of AI-generated deepfake scams. An advisory in late February 2024 specifically warned about digitally manipulated videos of public figures like Prime Minister Lawrence Wong, Senior Minister Lee Hsien Loong, and Deputy Prime Minister Gan Kim Yong. Total reported scam losses in 2024 hit SGD 1.1 billion, with 46,563 victims. The pitch is not new. The packaging is. It feels real because it looks real. The face moves, the voice wavers, the eyes blink. That is the point. It is also the trap.

Here is the uncomfortable truth: no single reliable visual tell lets you spot a deepfake by staring at it. The generative AI models that create these videos update continuously. A lip-sync artefact obvious in 2023 is gone by the time you see a video in 2025. Unnatural blinking, slightly waxy skin texture, a face too smooth around the hairline. These artefacts exist, and they are worth knowing as supporting clues. They are not a checklist. The moment a guide titled 'Look at the forehead creases' appears online, the next model update fixes the forehead creases. This is the AI realism trap: believing a video is authentic because it looks real, without considering that the thing you are looking at was generated by a machine trained on millions of hours of human faces. The question is not 'does this video look real?' The question is 'did this person actually say this?' To answer that, leave the video.

Forget the pixels. The only reliable defence is a behaviour called lateral reading. It is the opposite of the way you probably use the internet. Instead of staying on the page with the video, open a new tab. Leave the video. Search. Find the official social media account of the person in the video, or the press release from the body they supposedly represent, and check whether the endorsement exists there. For a Prime Minister or a Minister, the answer will be obvious: it will not be there. The Singapore government does not issue cryptocurrency endorsements through forwarded WhatsApp messages. This is verification at the speed of a competent internet user: seconds to open a tab, a minute to find the official source. In Singapore, the official source is often a website ending in .gov.sg. If the video is about an investment, check the Monetary Authority of Singapore's investor alert list. If it is a phone call from a bank officer, hang up and call the bank yourself. Lateral reading means you are not trying to judge the video in isolation. You check the claim against the world.

The Formula Never Changes

The most common deepfake scam in Singapore is the fake investment endorsement. The formula is almost always the same. You receive a message on WhatsApp or Telegram, often forwarded-as-received from a friend or family member. The message contains a link to a news article or a video. In the video, a public figure, sometimes a celebrity, sometimes a politician, appears to endorse an investment scheme. The scheme is extraordinary: guaranteed returns, no risk, a government-backed 'opportunity' that is actually a fake. The deepfake technology makes the person look and sound plausible. They might be sitting in a news studio, or in a hawker centre, or in what looks like an official office. The audio is synchronised, the head moves naturally, and the message is urgent: 'This is a limited window', 'Act now', 'Do not tell anyone until you have invested'. This is a deepfake impersonation scam, and it preys on the authority of a face you trust.

Tools the Police Provide

The Singapore Police Force's ScamShield app, available on iOS and Android, blocks known scam calls and SMS. The ScamShield WhatsApp bot at https://go.gov.sg/scamshield-bot lets you check suspicious messages and links. But the app is a stopgap. The real defence is your own suspicion when an investment sounds too good to be true.

Before the deepfakes, there were cheapfakes. A cheapfake is not AI-generated; it is media altered with simple tools. A video of a celebrity is slowed down, or a clip is taken out of context, or an old photo of a politician is reused in a new fake advertisement. In some cases, the scammers do not even try to create a new video. They take an old video of a celebrity speaking and add subtitles that put different words in their mouth. The lip movements do not match the subtitles, but the viewer is reading the subtitles, not watching the lips. This is a critical skill for AI-generated celebrity scam ad detection: do not watch a video with the sound off and subtitles on. Watch the person's mouth. Does the sound match the movement? Are the phonemes aligning with the bilabial sounds? If not, it is a fake. This is the 'lip-sync artefact' that deepfake detection guides, like Snopes' 2024 guide 'How to Spot a Deepfake', mention. It is not a perfect test. Good deepfakes are getting better. But for the sloppy, fast-generated scam ads that flood social media, it is often enough. Look for the cheapfake before you look for the deepfake.

While no single visual tell is reliable, some deepfake impersonation scam red flags are worth knowing, if only because they are not yet obsolete. Look for skin texture. Real video of a person's face has a certain noise to it, the grain of the camera sensor, the natural texture of pores. AI-generated faces often have a 'smoothness', a kind of waxiness that is too clean. Look at the eyes. The eyes are the hardest part to generate convincingly. Are the reflections in the eyes consistent with the light source in the scene? Do the pupils dilate? In many deepfakes, the blink rate is unnatural, either too frequent or not frequent enough. Look at the edges of the face, especially the hairline and the ears. Deepfakes often blur or 'melt' at the edges. Listen to the voice. Is the intonation off? Is there a slightly robotic timbre? These are the artefacts that are still hard to eliminate. But remember: these are supporting clues, not a checklist. A video that passes all of these tests can still be a deepfake. A video that fails one of them is definitely a fake. The absence of obvious artefacts is not proof of authenticity.

The primary vector for deepfake scams in Singapore is the forwarded-as-received message on WhatsApp and Telegram. Your aunt forwards you a video of the Prime Minister endorsing a scheme. You trust your aunt, so you trust the video. This is a failure mode called source confusion. You are judging the claim based on who sent it to you, not on where it came from. The 'forwarded' label exists for a reason. It means the person who sent it to you did not write it. They are just passing it along, often because they believe it is true. The verification bypass is when you see a video, doubt its authenticity, but then forward it to someone else with a comment like, 'Is this true?' You have just amplified a potential scam before you have verified it. The speed and frictionlessness of the share button have made you part of the problem. Do the lateral reading first. Before you forward anything that tells you to invest money or share your banking credentials, you have a responsibility to check. The Singapore Police Force has made this easy with the ScamShield WhatsApp bot, where you can paste a link and get a quick check. The bot is not a substitute for your own judgement. The best tool you have is a few seconds of hesitation.

When you do lateral reading, where should you go? In Singapore, you have several options. The government runs Factually, a fact-checking portal that publishes clarifications on false claims circulating locally. If a deepfake of a Minister is going around, Factually will likely have a clarificaton within hours. The verification speed of this source is a key strength. Black Dot Research, an independent fact-checker founded in 2021, also publishes local verifications. They are not affiliated with the government. Their website is blackdotresearch.sg. The Singapore Police Force Anti-Scam Command, formed in 2022, is the centralised police unit that investigates scams and publishes scam typologies and statistics. They also have a dedicated hotline: 1800-722-6688. If you encounter a video, check it against these sources. But do not just take one source's word for it. This is the single-source settlement failure. Finding one source that confirms your suspicion is not enough if that source is just regurgitating the same rumour. Look for two independent sources. A verdict without visible evidence is unverifiable by the reader. Source transparency, where the fact-checker names the original source and shows the evidence chain, is crucial. A fact-checker who says 'we have verified this' without showing their work is not helping you build your own skill.

The Laws Already in Place

Singapore has not been passive. The government has used the Protection from Online Falsehoods and Manipulation Act, or POFMA, to deal with deepfakes. On 3 June 2024, the POFMA Office issued the first deepfake-related POFMA direction, a Correction Direction, against Facebook and Instagram posts using a digitally altered video of Prime Minister Lawrence Wong. There were at least three such POFMA directions in 2024. The Elections (Integrity of Online Advertising) (Amendment) Act prohibits digitally generated or manipulated content that realistically depicts candidates during election periods. The Infocomm Media Development Authority has also proposed a Code of Practice for Online Safety that would require designated social media services to detect and remove AI-generated deepfake content.

How to Report a Deepfake

This is a fast-moving legal area, and the specific rules may change. But the existence of these laws tells you something important: creating or distributing these videos is a crime. If you see one, report it. The Singapore Police Force's iWitness portal is at www.police.gov.sg/i-witness. The ScamShield Helpline is 1799. The National Crime Prevention Council also has an anti-scam helpline at 1800-722-6688 and a website at www.scamalert.sg. Reporting is the right thing to do. Understanding the law is also part of media literacy.

You are scrolling through your social media feed, and you see a video of a celebrity, a well-known businessman, endorsing a cryptocurrency. The video has been shared by a friend. Here is what you do next. First, do not share it. Do not comment on it. Do not 'like' it. Every engagement, even negative engagement, tells the algorithm to show it to more people. This is algorithmic curation: the platform is optimising for engagement, not accuracy. You are feeding the machine. Second, do a reverse image search. If the video is a still image, use Google Images or Tineye to see where the image has appeared before. It might be an old photo that has been reused. Third, open a new tab and search for the name of the celebrity plus the name of the platform where you saw the video. You are looking for the official account. Does the celebrity have a verified account? Is the video there? If the official account has not posted it, it is fake. Fourth, check the ScamShield WhatsApp bot. Paste the link into the bot and see what it says. Fifth, if you suspect it is a scam, report it to the police. Finally, if you have already transferred money, call the ScamShield Helpline at 1799 immediately. The faster you report, the faster they can try to freeze the funds. Time is not on your side.

This is not just about a single video or a single scam. This is about a fundamental shift in how you must consume media. The entire premise of media literacy, as defined by the National Library Board's S.U.R.E framework, Source, Understand, Research, Evaluate, is that you cannot trust what you see. You have to judge it. The tools for judgement are changing. The S.U.R.E framework was developed before generative AI, but the core skill is still the same: can you determine whether a claim is true, and not merely whether it feels true? The AI realism trap is the modern version of this. It feels true because it looks real. Feelings are not a reliable guide. The evidence is. The evidence is only available if you do the lateral reading. This is a skill, and like any skill, it takes practice. Start with the videos you see from your own network. The next time someone forwards you a video of a politician or celebrity saying something outrageous, do not accept it. Verify it. Your grandmother, the one who forwarded you the scam, is not the one who will lose her life savings. The person who loses it is the one who does not check. Do not be that person.

Be honest about the trajectory of this technology. The deepfakes circulating today in Singapore, the ones that land people in the 13 cases reported to the police in H1 2024, are not perfect. They have tells. But the gap is closing. The generative AI models are being trained on more data, and they are being trained to fail the very tests that detection guides publish. The 'lip-sync artefact' described earlier is already being fixed. The unnatural blinking is being fixed. The waxy skin is being fixed. There is a future, not that far away, where a deepfake of a public figure will be indistinguishable from a genuine recording, even to a trained eye. When that day comes, the only thing that will save you is your behaviour. Your refusal to make a decision based on a video alone. Your commitment to lateral reading. Your willingness to pick up the phone and call the person. The technology will evolve, but the core literacy skill of questioning whether media is synthetic or authentic is permanent. The specific tools evolve. The skill does not. This is why the police and the fact-checkers teach lateral reading rather than just how to spot a deepfake. The 'spotting' has an expiration date. The 'reading' does not.

It is 2 am. Your phone rings. It is a video call from your son. His face looks right, his voice sounds right, but something is off. He says he has been in an accident and needs money urgently. He says he will transfer the details. He sounds desperate. What do you do? In this scenario, the Singapore Police Force has reported at least two cases in 2023 of deepfake-related extortion involving voice cloning of family members. This is the most dangerous kind of deepfake, because it bypasses your rational mind. Your parental instinct kicks in. But the rules are the same. Hang up. Call your son back on his regular number. Not the number that called you. A number you have used before. A number you have written down. If you cannot reach him, call your spouse, or the police. Do not send the money. This is the test of whether you have actually learned the lesson. It is easy to be skeptical of a video of Prime Minister Wong endorsing a cryptocurrency. It is much harder to be skeptical of your own child. The scammers know this. They use the AI realism trap to exploit your deepest fears. Verification speed in this scenario is critical. The longer you stay on the line, the more you are at risk. Get off the call. Verify. Then act.

The Singapore Police Force's ScamShield app is a useful tool. It is not a magic bullet. Its function is to block scam calls and SMS using a known-scam database. It was first released in 2020 and is available on iOS and Android. If you get a call from a known scam number, ScamShield will block it. If you get an SMS from a known scammer, it will filter it. But what about a deepfake video that arrives via WhatsApp on a number not yet in the database? The app cannot see video content. It cannot analyse a video for deepfake artefacts. Its user burden is low, you just install it and it runs in the background, but its tool accuracy is limited. It is a filter, not a detector. This is why the police have also launched the ScamShield WhatsApp bot at https://go.gov.sg/scamshield-bot. This is a more active tool: paste a link or a message and get a quick check against the police database. Even this bot is only as good as its database. It is a backstop, not a primary defence. The primary defence is still your own lateral reading. Do not over-trust the tool. The app is a useful layer of protection. It will not catch everything. And remember: the app is not a substitute for calling the ScamShield Helpline at 1799 if you have actually been scammed.

Spend a moment on Black Dot Research. This is an independent, Singapore-based fact-checking and media literacy organiation. They are not part of the government. They publish original verifications of claims circulating in Singapore, often in Chinese and English. They are part of a small but growing ecosystem of non-state fact-checkers. They do this work because they believe that a citizenry that can verify information is a citizenry that cannot be easily manipulated. Their approach is similar to the government's: check the source, look for evidence, publish findings with source transparency. But they are not constrained by government policy. They can say things that a government agency cannot. They can be more critical of a policy if the evidence warrants it. This matters. When you are doing your lateral reading, check both government and independent sources. A healthy information diet should include both. This is not because the government is lying. You need a diversity of sources to build an accurate picture. The National Library Board's S.U.R.E framework encourages this. Do not settle for a single source. Check Factually, check Black Dot Research, and check the mainstream news. If they all agree, you can be more confident. If they disagree, you have found something that needs more investigation.

There is a failure mode so common it deserves its own section: single-source settlement. This is when you find a single source that confirms what you want to believe, and you stop looking. The deepfake scam video says the investment is endorsed by the government. You go to a website that looks official, and it 'confirms' the endorsement. You stop there. You have been fooled. The official-looking website is run by the scammers. The 'confirmation' is part of the scam. This is a sophisticated version of the deepfake. It is not just the video that is fake. The entire ecosystem around it is fake. The fake news article, the fake testimonial from a fake investor, the fake website with a fake SSL certificate. This is why lateral reading is so important. Leave the original source entirely. Open a new tab and search for the name of the celebrity, the name of the scheme, and the word 'scam'. See what the independent sources say. This is effortful. It takes time. It is much easier to just believe. That ease is exactly what the scammers are selling. They are selling you a shortcut to wealth, and the price is your life savings. The next time you feel the urge to believe something that seems too good to be true, remember this: the easiest person to fool is the one who wants to be fooled.

What the S.U.R.E Framework Asks of You

The Media Literacy Council, or the MLC, is a Singapore charity that runs public education campaigns about media literacy. The MLC promotes the S.U.R.E framework, which stands for Source, Understand, Research, Evaluate. This is a teachable sequence embedded in Singapore's school curriculum and public libraries. It is a useful framework because it does not depend on any specific technology. It works whether you are looking at a deepfake video, a phishing email, or a forwarded message about a lottery win. Walk through it briefly. Source: Who is the author, and what is their agenda? Understand: What is the claim, and what does it mean? Research: What do other sources say? Evaluate: Is this claim true, or at least reliable? This is the framework to use when you encounter a deepfake. The source is the deepfake video itself. Who created it and why? The understanding is the claim being made. The research is the lateral reading you do to verify. The evaluation is the final judgement.

Why the MLC Matters Now

The MLC's role is to teach you this framework. They do not run the ScamShield app, and they do not write POFMA directions. They teach you how to think. In the age of generative AI, learning how to think is the most important skill you can have.

The POFMA directions of 2024 were not just about protecting individuals from losing money. They were about protecting the integrity of the political process. The deepfake of Prime Minister Wong was not trying to get people to invest in cryptocurrency, or at least, that was not the only thing it was trying to do. It was trying to sow confusion. It was trying to make people question whether the Prime Minister was really who he said he was. It was trying to create a world where nothing can be believed, and therefore anything can be true. This is the real danger of deepfakes. The 'liar's dividend' is the benefit that a liar gets from a world where nobody can trust anything. When you can no longer believe your own eyes, you can no longer believe anything. And when you can no longer believe anything, you will believe anything. This is why Singapore has been so proactive in using POFMA to take down deepfakes of its leaders. It is not just about the individual video. It is about the precedent. It is about establishing that this content is illegal and will not be tolerated. The Elections (Integrity of Online Advertising) (Amendment) Act, with its deepfake provision, is a direct response to this threat. During an election, the stakes are even higher. The law is a tool, but it is a blunt tool. The sharpest tool is you.

There is a cost to all of this. The cost is speed. When you do lateral reading, you are trading speed for accuracy. You are sacrificing the immediate gratification of the 'like' button for the long-term security of knowing the truth. This is a trade-off that many people are not willing to make. They would rather share a piece of misinformation quickly and be wrong, than take the time to verify and be right. This is the verification bypass mentioned earlier. It is the cause of so many of the problems seen online. The Singapore Police Force, through its Anti-Scam Command, has tried to make the verification process as fast as possible. The ScamShield WhatsApp bot can give you an answer in seconds. Factually often publishes its clarifications within hours of a false claim going viral. The verification speed of these official sources is a key part of the defence. Even a 30-second delay is too long for some people. Build the verification habit. The next time you see a video of a celebrity or a politician endorsing an investment, do not share it. Take a breath. Open a new tab. Check the source. It will take you 30 seconds. It could save you your life savings. The cost of being right is a few seconds of your time. The cost of being wrong is everything.

What is coming next? A few trends are likely to continue. First, the deepfakes will get better. The artefacts visible today will be gone within a few years. Second, the scams will become more targeted. Instead of a mass broadcast to millions of people, scammers will use AI to create personalised deepfakes for specific individuals. They will scrape data from social media to learn about your interests, your family members, your daily routines, and then they will use that data to create a video uniquely convincing to you. This is the 'digital shadow' concept: the data you leave behind as you browse. Third, the official response will also get better. The Singapore Police Force will continue to develop its detection capabilities. The IMDA will likely finalise its Code of Practice for Online Safety, requiring platforms to do more to detect and remove deepfakes. But the arms race will continue. There is no technological solution to this problem. The only solution is human: a population media-literate enough to question what it sees. The Singapore Police Force reported 13 deepfake cases in H1 2024. That number is small compared to the 46,563 total scam victims. It is growing. The future of deepfake scams is not a question of 'if' but 'how many' and 'how sophisticated'. The only question is whether you will be one of the victims.

If you have already fallen for a deepfake scam, do not panic. Time is of the essence. The first thing you should do is call the ScamShield Helpline at 1799. This is the Singapore Police Force Anti-Scam Command hotline. Report what happened. Give them the details. The faster you call, the better your chances of getting your money back. Second, call your bank. Tell them you have been scammed and ask them to freeze your accounts. The bank may be able to reverse the transaction if it was made recently. Third, go to the police and file a report. You will need a police report number for your bank and for any other institutions you need to contact. Fourth, report the scam to the platform where you saw the video. This is important for algorithmic curation: you are helping to flag the content so that it is shown to fewer people. Finally, tell your friends and family. Warn them. The scammers often target the same people multiple times. If you have been scammed, your name is on a list, and they will try to scam you again. Do not be embarrassed. This can happen to anyone. The Singapore Police Force reported that in 2024, victims lost a total of SGD 1.1 billion to scams. You are not alone. There is no shame in seeking help.

This is not just a Singapore problem. The same deepfake technology is being used in countries all over the world. In the United States, deepfakes have been used to influence elections. In Europe, deepfakes have been used for financial fraud. The Singapore Police Force's advisory of 29 February 2024 mentioned that the impersonated professionals in these scams were not just Singaporean politicians like Prime Minister Lawrence Wong and Senior Minister Lee Hsien Loong, but also global figures like Elon Musk, Warren Buffett, and Jeff Bezos. This is a global problem, and it requires a global response. But the response has to start with you. You cannot rely on the tech companies to solve this problem. Their algorithmic curation systems spread the content. You cannot rely on the government alone. They are always playing catch-up. The only reliable defence is a population that is media-literate. This is the mission of the Media Literacy Council and the S.U.R.E framework. It is a long-term project. It is not glamorous. It is the only thing that will work.

Boil this down to the essentials. You are sitting in a hawker centre in Toa Payoh, or at a desk in a Raffles Place office, scrolling through your phone. You see a video of a celebrity endorsing an investment. What do you do? Remember your training. Do not trust the video. Open a new tab. Search for 'celebrity name' + 'investment scam'. Find the truth. Report the video to ScamShield. Move on with your day. That is the entire page in a nutshell. It is not about being paranoid. It is about being prepared. It is about having a habit of verification so strong that it is automatic. The AI-generated content is not going away. It is going to get better, cheaper, and more prevalent. The only question is whether you will be able to tell the difference between a real video and a fake one. The answer, if you follow the advice here, is yes. You will be able to tell. Not because you have some magical ability to see the pixels, but because you have a simple, reliable, and effective method. Leave the video. Check the source. Decide. That is the whole secret.

The honest caveat is this: no method is perfect. Lateral reading will not catch every deepfake. There will be videos so good, so indistinguishable from reality, that they will fool even the most careful eye. The verification speed of fact-checkers like Factually and Black Dot Research is fast. It is not instant. There will be a window of time, maybe hours, maybe days, where a deepfake circulates before it is debunked. During that window, you could be fooled. The only defence against this is to adopt a default stance of skepticism. Not cynicism. Skepticism. Do not believe that everything is fake. But do not believe that everything is real. The burden of proof is on the claim, not on the skeptic. This is a hard standard to maintain. It is the only one that works. The AI realism trap is real. The technology is remarkable. You are more remarkable. You have the ability to think, to question, and to verify. Do not give that up. Use it. Every time you are presented with a claim that seems too good to be true, or too outlandish to be false, slow down. Open a new tab. Do the work. The future of our information ecosystem depends on it.

The Singapore Police Force reported 13 deepfake cases in H1 2024. This is a specific, verifiable number from an officially published advisory, and it is the kind of concrete detail that a generic 'how to spot a deepfake' listicle would not know to include, or would not bother to verify, and it is what anchors the entire page in the local Singaporean context rather than in abstract theory.