Key Takeaways
- Singapore reported S$1.1 billion in scam losses in 2024, with phone and digital scams accounting for the majority
- AI voice cloning technology now enables convincing impersonation of family members, bank officers, and government officials
- Between 2022 and 2023, deepfakes detected in the Asia-Pacific region rose 1,530%, according to Sumsub
- Singapore's ScamShield app filters known scam calls and messages, but novel AI-generated approaches are harder to catch
- In March 2025, SPF, MAS and CSA issued a joint advisory warning businesses about scams involving digital manipulation, including deepfake impersonation
The Facts
Singapore is one of the world's most connected, digitally literate societies — and it lost S$1.1 billion to scams in 2024. The figure, reported by the Singapore Police Force, represents a deeply uncomfortable data point for a city-state that has invested heavily in public digital literacy campaigns and anti-scam infrastructure.
The profile of scam losses has shifted. Traditional phone scam typologies — bank impersonation, government official impersonation, investment fraud — have been augmented by AI-assisted variants that defeat the pattern-recognition awareness training that anti-scam campaigns rely on. Voice cloning technology, trained on as little as a few seconds of audio extracted from social media posts, WhatsApp voice notes, or LinkedIn video bios, can generate real-time synthesised audio indistinguishable from the target's actual voice.
Voice cloning makes it possible to impersonate parents, siblings, or employers — creating urgent emotional scenarios (a family member in hospital, a boss needing an emergency fund transfer) that bypass rational scam assessment in the same way genuine emergencies would. The temporal pressure of these scenarios — "I need the money in the next hour" — deliberately prevents the victim from taking the verification steps they might otherwise consider.
Sumsub's identity fraud research recorded a 1,530% rise in deepfakes detected across Asia-Pacific between 2022 and 2023, as generation tools became cheaper and more accessible.
Technical Deep-Dive
AI voice cloning for fraud operates through a three-stage pipeline. Audio collection gathers target voice samples from publicly available sources — social media video posts, podcast appearances, YouTube recordings, LinkedIn video bios, or voice notes previously sent to the attacker via social engineering. Training or fine-tuning a voice synthesis model on these samples takes minutes to hours with current consumer tools, producing a model that generates real-time speech in the target's voice from text input.
The real-time synthesis capability is the critical differentiator from earlier deepfake techniques. Older approaches required pre-recorded clips — limited, detectable, and unable to respond dynamically. Current real-time voice synthesis tools can be operated by an attacker during a live phone call, typing responses that are spoken by the synthesised voice to the victim in real time. The victim hears their family member's voice responding naturally to their questions.
Detection is technically challenging. Anti-voice-cloning tools exist that analyse acoustic patterns for synthesis artefacts, but these require the potential victim to proactively scan an incoming call — a step most people do not take during an emotional phone call scenario.
The ASEAN Perspective
Singapore's S$1.1 billion in scam losses includes investment scams, and AI tools make it cheaper to build convincing fake adviser personas — social media profiles with AI-generated profile photos, AI-written market commentary, and AI voice call follow-ups.
The ScamShield app — Singapore's national anti-scam mobile application — provides SMS and call filtering based on known scam numbers and text patterns. It is less effective against AI-generated calls from dynamically generated numbers with synthesised voices that do not match any known scam pattern signature.
In March 2025, SPF, MAS and CSA issued a joint advisory warning businesses about deepfake impersonation scams and urging verification through established internal channels. Out-of-band checks of that kind — confirming a request through a known number or channel rather than the one that made it — are more durable than technical signature-matching against AI-generated voice fraud.
RECATOOLS Verdict
The S$1.1 billion figure demands honest assessment: traditional anti-scam strategies built around pattern recognition and public awareness are encountering a capability ceiling against AI-generated social engineering. The scam ecosystem has adopted AI faster than the defence ecosystem.
The durable response is architectural: verification procedures that are independent of the communication channel being manipulated, rather than improved pattern recognition within the same channel.
Sources
- Singapore Police Force Annual Scam Statistics 2024
- Sumsub identity fraud research, 2023
- iProov Biometric Identity Threat Intelligence 2026
- SPF, MAS and CSA Joint Advisory on Scams Involving Digital Manipulation, March 2025
FAQ
How much did Singapore lose to scams in 2024? S$1.1 billion, according to the Singapore Police Force's annual scam statistics report.
How does AI voice cloning enable scams? Real-time voice synthesis creates convincing audio impersonations of family members or officials from small samples of publicly available audio — enabling emotional manipulation scenarios that bypass rational scam assessment.
What is ScamShield? Singapore's national anti-scam mobile application that filters known scam SMS and calls. Effective against known-pattern scams; less effective against AI-generated novel social engineering.
What verification should Singapore residents use for urgent money requests? Call back to a pre-saved number for the person or institution — not the number that called you — to independently verify any urgent financial request regardless of how convincing the caller sounds.
Is AI voice cloning legal in Singapore? Using AI voice cloning for fraudulent purposes is criminal under Singapore's Computer Misuse Act and Cheating provisions. The technology itself is not illegal; its fraudulent application is.
Correction, 18 October 2026: An earlier version misdescribed a March 2025 joint advisory by the Singapore Police Force, MAS and CSA, attributed a deepfake statistic to Positive Technologies rather than Sumsub. Unsourced statements about victim reports and Singapore's per-capita losses have been removed.
View author profile → · Editorial policy
