Home › Insights › Threat Landscape
Threat Landscape

AI Voice Cloning and Deepfake Scams Are Hitting Singapore Businesses: The Threat Behind the Headlines

AI voice cloning and deepfake video calls are being used to impersonate executives and authorise fraudulent payments in Singapore, and most businesses still have no process to verify who is really on the call.

By Infracom Team10 November 20267 min read
Watch the short explainer, then read on. More on our YouTube channel.

AI voice cloning and deepfake video calls are now a live fraud channel against Singapore businesses, not a theoretical risk. Criminals use a few seconds of scraped audio or video to fabricate an executive's voice or face in real time, then use that fabrication to instruct a finance team to move money. The gap is not detection technology — it is the absence of a verification process that does not rely on recognising a face or voice on a screen.

What is actually happening

Singapore authorities have moved from warning the public about scam calls to warning businesses about boardroom-grade impersonation. On 12 March 2025, the Singapore Police Force, Monetary Authority of Singapore and Cyber Security Agency of Singapore issued a joint advisory on scams involving digital manipulation, in which Artificial Intelligence is allegedly used to create or manipulate synthetic media, and set out precautionary measures for businesses to adopt.

The cases behind the advisory are not abstract. CSA has referenced an incident in which an employee of a multinational firm was tricked into sending $25m to scammers, after receiving a deepfake video that appeared to be his chief financial officer giving payment instructions. CSA has also noted that other known scams have used deepfake videos of Prime Minister Lawrence Wong and senior minister Lee Hsien Loong.

Singapore-specific cases keep surfacing through 2025 and 2026. In one, a finance director at a multinational in Singapore transferred more than US$499,000 after a Zoom call in which the CEO and other executives were impersonated using deepfakes. It was only when the scammers pushed their luck that the fraud unravelled: when the criminals asked for a further US$1.4 million, the victim alerted the company's bank, which notified Singapore's Anti-Scam Centre, and working with Hong Kong police, authorities froze and recovered the funds that had been moved.

A separate case involved a much larger loss. The Singapore Police Force disclosed that a businessman lost roughly S$4.9 million after a scam that began with a WhatsApp message, which escalated into a fabricated video call and an email that included what appeared to be an official government-issued letter of guarantee bearing PM Wong's signature, stating the funds would be reimbursed by the Singapore Government within 15 business days. Police later released footage from a separate, related scam operation showing a fabricated Zoom conference: the video conference was fabricated, with some portions edited using deepfake AI technology, showing a fabricated discussion about the situation in the Straits of Hormuz, before a deepfake fabrication that can be sophisticated and difficult to differentiate from authentic content.

The scale of the shift shows up in the national numbers too. Police figures show Singapore recorded 37,308 scam cases in 2025, down from 51,501 the year before, with scams still accounting for 88.9% of all scam and cybercrime cases reported, and reporting indicates AI-enabled impersonation, often through video and voice cloning, is the fastest-growing vector. Overall case numbers falling while this specific vector grows is the signal operators should take seriously: the threat is concentrating, not dissipating.

What regulators and frameworks expect

Three Singapore bodies have now put this threat in writing, and each expects a different layer of control.

← Swipe to compare →

DimensionSPF / MAS / CSA Joint AdvisoryCSA Guidelines on Securing AI SystemsMAS Guidelines on AI Risk Management
Issued12 March 202515 October 2024, with an agentic AI addendum7 October 2026, effective 7 October 2027
Who it is aimed atAll businesses and the publicSystem owners adopting or building AIAll financial institutions, proportionate to size
Core expectationBusinesses should adopt precautionary measures against scams using AI-manipulated synthetic mediaSystem owners should secure AI throughout its lifecycle, protecting against classical cybersecurity risks and novel risks such as adversarial machine learningFinancial institutions must manage AI risk while tailoring approaches to the scale and nature of their AI use
SourceCSA advisory AD-2025-005CSA publicationsMAS media release

CSA's Guidelines on Securing AI Systems take a lifecycle view across Planning and Design, Development, Deployment, Operations and Maintenance, and End of Life, and are supported by a Companion Guide that is a community-driven resource curating practical measures, security controls and best practices, and references resources such as the MITRE ATLAS database and OWASP Top 10 for Machine Learning and Generative AI. CSA has since extended this with an addendum: Securing Agentic AI, which provides practical guidance for system owners to secure Agentic Artificial Intelligence systems.

On the financial sector side, MAS has moved from voluntary good practice to supervisory expectation. Its new Guidelines require financial institutions to ensure their board and senior management provide effective oversight of AI risks, including by setting clear roles and responsibilities, risk appetite, and risk management frameworks, policies and procedures, with controls covering data governance, testing, human oversight, cybersecurity, monitoring and change management, with the extent and sophistication of controls depending on an institution's risk exposure. These Guidelines take effect from 7 October 2027, giving institutions a defined runway to close gaps rather than an open-ended expectation.

Alongside these, PDPC's role matters for any business using likenesses, voice samples or synthetic data in AI tooling. Its published guidance on this subject sets out that organisations generating synthetic data should follow good practices and risk assessments to reduce residual risks such as re-identification — a relevant discipline wherever employee or customer voice and image data feeds AI-enabled verification or detection tools.

How to prepare and prevent

None of the advisories above ask businesses to detect deepfakes with certainty — none of the agencies claim that is reliably possible yet. What they ask for instead is a verification process that does not depend on recognising a face or voice.

  1. Out-of-band verification for payment instructions. Any request to move funds, change bank details, or approve an unusual transaction that arrives by video call, voice call or message should be independently confirmed through a separate, pre-agreed channel — not by calling the number given in the request itself.
  2. A documented callback and escalation procedure. Name who can authorise high-value or urgent payments, require a second approver, and set a threshold above which no single call or message is sufficient authority.
  3. Treat urgency as a red flag, not a reason to move faster. Deepfake scams rely on time pressure and authority bias; a mandatory cooling-off step for unusual requests blunts both.
  4. Maintain an AI asset and tooling inventory. If your business uses AI for customer verification, voice authentication, or any agentic workflow, you need to know what it is, where it sits in your environment, and who owns its risk — the starting point CSA's lifecycle guidance and MAS's new Guidelines both assume is already in place.
  5. Train staff on the specific tell-tale signs flagged in police advisories, such as audio that does not sync with lip movement, background artefacts, or audio broadcast through a single account rather than individual participants, while treating these as one screen among several rather than the decision.
  6. Report promptly. Early escalation to your bank and the Anti-Scam Centre, as in the Singapore case where funds were frozen through cooperation with Hong Kong police, is what actually recovers money — speed matters more than certainty.
  7. Close the data exposure. Limit how much executive voice and video content is publicly available, and review what your organisation publishes that could train a convincing clone.

For reporting channels and current scam typologies, businesses should monitor the Singapore Police Force's advisories, ScamAlert.sg, and CSA's alerts page directly, rather than relying on second-hand summaries.

Structurally, the strongest defence is to put this inside a certified cybersecurity governance programme rather than treating it as a one-off staff briefing. CSA's Cyber Trust Mark, published as a national cybersecurity certification developed by CSA and published as Singapore Standard SS 712:2025, is organised in five tiers where Cyber Essentials sets a baseline of operational cyber hygiene while Cyber Trust recognises governance and risk management maturity. Notably, the trust mark includes a domain focused on the security of AI usage within organisations, as part of a broader set of cybersecurity preparedness measures — which is exactly the gap that deepfake and voice-cloning fraud exposes. For SMEs without a mature security function, Cyber Essentials readiness is usually the right starting point before stepping up.

How Infracom helps

Infracom governs, assesses and secures the AI systems and processes your business already relies on — we do not build or train AI. Where deepfake and voice-cloning fraud exposes gaps in verification procedures, payment controls, or AI tooling oversight, our GRC advisory work maps your exposure against the CSA, SPF and MAS advisories above, supports Cyber Trust Mark or Cyber Essentials readiness, and, where a business needs ongoing senior oversight of AI risk without a full-time hire, our fractional Chief AI Officer service (/grc/vcaio/) provides that governance function on a fractional basis, delivered by an all-Singaporean SQEP team at SME-accessible rates.

The AI Governance Pack holds the AI inventory, the risk and impact assessment, the ISO/IEC 27001 to 42001 gap map, the policy set and the board paper as a workbook and document set.

Sources (11)
  1. CSA, SPF, MAS Joint Advisory on Scams Involving Digital Manipulation
  2. SPF footage release on Zoom deepfake impersonating senior government officials
  3. CSA Guidelines and Companion Guide on Securing AI Systems
  4. CSA Addendum on Securing Agentic AI
  5. MAS Sets Out Supervisory Expectations on Responsible AI Adoption by Financial Institutions
  6. PDPC Guide on Synthetic Data Generation
  7. Singapore businessman loses $3.8M in deepfake Zoom call posing as PM Lawrence Wong (VnExpress)
  8. Singapore Issues Multi-Agency Advisory On AI, Deepfake Scams (Vixio)
  9. The Cloned Voice at the Top: Deepfake Executive Impersonation
  10. Singapore Cyber Trust Mark: Complete Guide for Businesses
  11. CSA Cyber Trust mark publication (SS 712:2025)

Get Infracom Insights by email

Practical cybersecurity governance and regulatory updates for Singapore and Australia. No more than a few emails a month; unsubscribe any time.

Infracom Consultancy Integration Pte Ltd

Your one-stop IT & cybersecurity partner — Singapore HQ since 2008, expanding to Australia in 2026.

506 Chai Chee Lane

Singapore 469026

Services
IT InfrastructureCloud SolutionsCybersecurityManaged ServicesData Center SolutionsIT Consulting
Consulting
SQEPGRCVAPT
GRC Services
Fractional CISOFractional CAIOData Protection OfficerInternal & Supplier Audit
Approach
MethodologyAbout UsInsightsStoreNewsletterPrivacy Policy
Hours

Mon – Fri
9AM – 6PM SGT

Follow us
© Infracom Consultancy Integration Pte Ltd. All rights reserved.Privacy Policy