The key to spotting a deepfake? Look beyond the audio or visual cues

In July last year, retiree Martin Ng was browsing music videos on the YouTube mobile application when a familiar face appeared on his screen.

In a video ad, “Senior Minister Lee Hsien Loong” was promoting what appeared to be an attractive cryptocurrency investment scheme.

At first glance, it looked convincing and credible, he says. 

But as Ng continued watching, “I had a strong gut feeling that it was a scam”. The 69-year-old recalls reading about deepfakes impersonating political figures in the news.  

What gave it away was the voice, Ng says. The “SM Lee” in the video looked real, but sounded robotic with unnatural intonations – which Ng perceives as an “obvious” tell-tale sign that the clip was a deepfake. 

Deepfakes are image, video, and audio content that have been created or manipulated by artificial intelligence.

Though he did not fall for the scam, Ng is worried it will be much harder to spot deepfakes in future. 

“I know how quickly the technology has improved (since I saw that video a year ago),” he says. Ng adds that the deepfake scam videos involving senior government officials, which he came across in recent months, are now “even more realistic”.

Ng’s concerns reflect a growing challenge in the fight against scams: AI is making it easier for scammers to create convincing fake videos, cloned voices and realistic impersonations of public figures. 

Global policing agency Interpol, headquartered in France, noted the rapid growth of deepfake and AI-enabled cybercrimes in a report published June 2026. 

From February to June 2024, there was a 600 per cent rise in discussions about deepfakes on cybercriminal forums and channels on the Telegram messaging app that are popular among South-east Asian threat actors, the report says.

The study dives into cyber threats and trends in the Asia and South Pacific region from January 2024 to March 2025.

According to Singapore Police Force (SPF) figures, there were 37,308 scam cases in 2025, with victims losing $913.1 million in total. SPF does not track AI-enabled scams as a separate category.

AI tools are now increasingly accessible, allowing threat actors to carry out scams at “greater speed, scale and sophistication”, says Luke Ho, director of National Cyber Threat Analysis Centre at the Cyber Security Agency of Singapore (CSA).

Harder to spot, easier to trust

For Ng, what gave the deepfake away was the robotic voice.  

Some visual and audio inconsistencies – such as unnatural facial complexions or out-of-sync lip movements – may provide clues on whether a video is real, says CSA’s Ho. 

But with AI becoming more advanced, he cautions against relying solely on visual or audio cues to determine whether a video is AI-generated.

Instead, Ho urges the public to focus on the broader context and adopt a “healthy dose of scepticism”.

Some questions to consider include:

These are often more reliable warning signs of a scam than imperfections in the video itself, says Ho.

Trust as a weapon

Scammers do not rely on deepfakes alone, Ho adds. Often, they are combined with tactics such as authority and familiarity to build trust.

One such case occurred in March 2025, when a finance director at a multinational firm nearly lost close to US$500,000 (S$646,000) in company funds. He believed he was attending a video conference with the company’s chief financial officer and other senior executives to discuss a restructuring of the firm’s regional business.

In reality, the participants were impersonated using deepfake technology. 

The victim was instructed to transfer over US$499,000 from the company’s corporate bank account to another local corporate bank account. He did as he was told, and only became suspicious after receiving a request for a further transfer of US$1.4 million. 

The victim alerted the company’s corporate bank, leading to the scam being uncovered. The initial funds had already been moved to bank accounts in Hong Kong, but were later recovered through collaboration between Singapore and Hong Kong authorities.

Once trust is established, scammers often introduce urgency or authority to pressure victims into acting before they have time to pause and verify the request, Ho explains. 

“It is therefore important that we stop and check with official and trusted sources before taking action,” he adds.

Since coming across the deepfaked video last July, Ng has become more cautious. 

For example, earlier this year, when he came across YouTube videos on the Budget announcements, Ng cross-checked the information against the official Budget website rather than taking the information at face value.

A simple rule Ng follows when it comes to verifying information online: “(Double check) against official government websites or ask my daughter.”

This is part of a series titled “Act against scams”, in partnership with the Singapore Police Force and the National Crime Prevention Council.