A finance employee at the Hong Kong office of British engineering firm Arup joined what looked like a routine video conference with the company's UK-based Chief Financial Officer and several other "colleagues." Every face on that call, except his own, was an AI-generated deepfake. (Hong Kong Police briefing of 2 February 2024, reported by CNN on 4 February 2024; Arup confirmed it was the victim on 17 May 2024 - CNN Business.)

That was the wake-up call for Asia-Pacific finance and treasury teams, including in Singapore. By 25 March 2025, a finance director at a multinational firm in Singapore had wired US$499,000 (about S$670,000) after a near-identical ruse: a WhatsApp approach from the company's "CFO", a Zoom call with the "CEO" and other "officers" deepfaked, and a purported lawyer asking him to sign an NDA. (SPF, reported by HRD Asia, 9 April 2025; MAS Information Paper, September 2025.) (Singapore Police Force / MAS / CSA Joint Advisory, 12 March 2025.)

If your business pays vendors, runs payroll, or moves money on instruction from senior staff - which is to say, if your business is a business - here is what is happening, what the law in Singapore says, and where Cyber, Crime/Fidelity, and Social Engineering insurance does and does not respond.

The Arup case, reconstructed

The attack on Arup's Hong Kong office is the case MAS's deepfake paper and CSA's advisory both cite, and the most useful case study for a small business, because almost nothing about it required exotic technology.

In mid-January 2024, an employee in the Hong Kong finance department received an email purportedly from Arup's UK-based CFO referring to a "secret transaction" that needed to be carried out. The employee initially thought it was phishing. (South China Morning Post, 17 May 2024.)

Then came the video conference. On the call, the employee saw and heard what looked and sounded like the CFO and several other senior colleagues. Hong Kong Police Senior Superintendent Baron Chan Shun-ching told public broadcaster RTHK that "in the multi-person video conference, it turns out that everyone [he saw] was fake." (CNN, 4 February 2024.) The employee was instructed to make 15 transfers to five Hong Kong bank accounts, totalling HK$200 million. He realised he had been scammed only after following up with Arup's UK headquarters.

Arup told the Financial Times: "We can confirm that fake voices and images were used. Our financial stability and business operations were not affected and none of our internal systems were compromised." Its global Chief Information Officer, Rob Greig, told Dezeen: "Like many other businesses around the globe, our operations are subject to regular attacks, including invoice fraud, phishing scams, WhatsApp voice spoofing, and deepfakes. What we have seen is that the number and sophistication of these attacks has been rising sharply in recent months." (Dezeen, 17 May 2024.)

Three points matter for you:

  • The attack did not breach Arup's network. No malware, no stolen passwords, no access to Arup's IT environment. The attackers exploited human trust, not technical vulnerability. The Professional Risk Managers' International Association case study notes that all of Arup's traditional cybersecurity layers - firewalls, MFA, endpoint protection - were operating effectively throughout. (PRMIA Case Study on Arup.) Cyber controls that defend against intrusion will not stop a deepfake video call.
  • No arrests, no recovery. As of early 2025, Hong Kong Police investigations remained ongoing and no perpetrator had been publicly identified or stolen funds recovered, according to the most current public reporting. (PurpleSec Arup case file.)
  • The mechanics are reproducible. The attackers built convincing avatars from publicly available material - LinkedIn videos, conference recordings, interviews. McAfee's Beware the Artificial Impostor research (May 2023) found that three seconds of audio was enough for a free voice-cloning tool to produce a clone with an 85% voice match to the original. (McAfee.) OpenAI's Voice Engine, announced 29 March 2024, requires only a 15-second audio sample.

What is happening in Singapore

Singapore's regulator, police force, and cybersecurity agency have all gone on the record that this attack pattern is now operating against companies based here.

On 12 March 2025, the Singapore Police Force, Monetary Authority of Singapore, and Cyber Security Agency of Singapore issued a joint advisory titled Scams Involving Digital Manipulation. The advisory describes "scams involving digital manipulation, in which Artificial Intelligence (AI) is allegedly used to create or manipulate synthetic media (i.e. deepfakes)." It describes the now-familiar pattern: an unsolicited WhatsApp message from a "high-ranking executive," followed by an invitation to a live-streamed Zoom video call where digitally manipulated participants instruct the victim to transfer funds. (SPF/MAS/CSA Joint Advisory, 12 March 2025.)

The headline Singapore case followed two weeks after that advisory and was reported by the SPF in April 2025. On 24 March 2025, the finance director of a multinational firm in Singapore was contacted on WhatsApp by someone purporting to be the company's CFO. On 25 March, the director joined a Zoom call featuring the "CEO" and other "officers" - all deepfaked. A separate "lawyer" called him and asked him to sign a non-disclosure agreement. The director transferred US$499,000 from the company's HSBC account to a Singapore mule account, from which US$494,000 was onward-transferred to Hong Kong accounts. He realised it was a scam only when an additional US$1.4 million transfer was demanded. The Singapore Anti-Scam Centre and Hong Kong's Anti-Deception Coordination Centre worked together to seize the full amount in Hong Kong by 28 March. (HRD Asia, citing SPF; Reality Defender summary of MAS report.)

Six months later, on 18 September 2025, MAS published an information paper titled Cyber Risks Associated with Deepfakes (MAS/TCRS/2025/06). It is intended for financial institutions but reads as a near-checklist for any business at risk. The paper identifies three core deepfake threat vectors: defeating biometric authentication, social engineering and impersonation scams, and misinformation/disinformation aimed at investor confidence. It cites the Arup case and the Singapore US$499,000 case as examples. (MAS Information Paper, September 2025.)

The broader Singapore scam picture frames why all of this matters at the SME level:

  • The SPF's briefs put scam losses at at least S$1.1 billion in 2024 (up 70.6% from S$651.8 million in 2023), S$913.1 million in 2025 (down 17.9%) and about S$410.6 million in the first half of 2026 (down a further 17.9% on the same period of 2025). (SPF Annual Brief 2024; Annual Brief 2025; Mid-Year Brief 2026.) (SPF Annual Scams and Cybercrime Brief 2024.)
  • Business Email Compromise losses were S$88.5 million in 2024, fell 60.1% to S$35.3 million in 2025, then rose 193.1% to S$57.3 million in the first half of 2026 alone, from S$19.5 million in the same period of 2025, across 262 reported cases; BEC sat in the top five scam types by amount lost in every one of those periods, and the SPF's mid-year brief now lists "use of digital manipulation" among the ways the impersonation is done. (SPF Mid-Year Scam and Cybercrime Brief 2026; Annual Brief 2025.)
  • The largest single BEC case in 2024 hit a Singapore commodities firm. On 19 July 2024, the firm transferred US$42.3 million (about S$57.2 million at the time) to what staff thought was a long-time supplier; the attackers had spoofed the supplier's domain by replacing an "i" with an "l". The SPF Anti-Scam Centre, working with INTERPOL and Timor-Leste authorities, had US$39 million detected and frozen in Timor-Leste the day after the report was lodged, and the Timor-Leste authorities arrested seven suspects. (SPF News Release, 3 August 2024.)
  • The 2025 picture eased: the SPF Annual Scam and Cybercrime Brief 2025 reported 41,974 scam and cybercrime cases (down 24.8%) and S$913.1 million in scam losses (down 17.9%), and BEC losses fell 60.1% before rebounding in 2026. (SPF Police Life summary.)

Deepfakes of senior Singapore Government officials have been used in retail investment scams since at least December 2023 - including videos of then-DPM Lawrence Wong and former PM Lee Hsien Loong endorsing fake investment products. (CSA Advisory AD-2024-006.) On 7 March 2025, PM Wong publicly warned that deepfakes of him were being used to sell crypto schemes and PR application services. (Yahoo News / Bloomberg, 7 March 2025.) These campaigns are aimed at retail investors, not SMEs directly, but they tell you the technology is in active use against Singapore-based audiences.

How the attack actually works

The technical floor has dropped. None of this requires a state actor.

Source material. Attackers harvest video and audio from LinkedIn, YouTube, conference recordings, internal town-halls reposted publicly, and media interviews. McAfee's 2023 research found that three seconds of audio produced a clone with an 85% voice match (McAfee); OpenAI described Voice Engine on 29 March 2024 as generating speech from "a single 15-second audio sample" (OpenAI); PurpleSec's Arup case file puts the working requirement at 20 to 30 seconds and lists LinkedIn profile videos, conference presentations and recorded video meetings as the harvest (PurpleSec).

The "live deepfake." PRMIA's case study calls the Arup call one of the first publicly confirmed cases of real-time deepfake impersonation driving a major corporate fraud, and MAS says the fraudsters built the deepfake videos from publicly available footage of the CFO and his colleagues. The Arup attackers populated a multi-participant call with several deepfaked "executives" simultaneously - not because each was unique, but because the social pressure of seeing several colleagues agreeing in real time overrides individual scepticism.

Voice cloning is a risk to voice biometrics. Some banks and corporate phone systems use voice-print authentication, and MAS's deepfake paper says audio deepfakes, like video and image ones, have the potential to defeat biometric authentication. The CrowdStrike 2025 Global Threat Report recorded a 442% rise in voice phishing (vishing) operations between the first and second halves of 2024. (CrowdStrike 2025 Global Threat Report findings.)

Synthetic identity for KYC bypass. The same technology is being used to defeat onboarding KYC - submitting deepfake selfies and document photos. At the same 2 February 2024 briefing, Hong Kong police said that "on at least 20 occasions, AI deepfakes had been used to trick facial recognition programs by imitating the people pictured on the identity cards." (CNN, 4 February 2024.)

Delivery vectors. The channels in the reported cases are spear-phishing email, WhatsApp, Microsoft Teams, Zoom and voice calls, and the Arup and Singapore March 2025 cases combined two: an opening message (an email in the Arup case, WhatsApp in the Singapore case), then the video call itself.

The "look for weird artifacts" advice is not enough on its own. Older guidance told staff to look for unnatural blinking, lighting mismatch, or lip-sync drift. The SPF/MAS/CSA advisory of 12 March 2025 still lists tell-tale audio-visual signs to check, and CSA's own advisory says "deepfake detection tools for general consumer use are still nascent"; treat the visual check as one screen among several, not the decision. (CSA Advisory AD-2024-006.)

The "callback" myth. The standard advice - "if in doubt, hang up and call the person back on their normal number" - works only if the attacker has not also compromised the contact route. Attackers have been seen using a fake WhatsApp profile with the executive's photo, then routing the "call back" to themselves, or explaining that they are calling from a different number because the matter is confidential (the Ferrari case, below). The verification has to happen on a channel and number you established before the request, not one provided in the request.

Other deepfake fraud cases worth knowing

WPP, May 2024 (attempt, unsuccessful). Fraudsters set up a fake WhatsApp account using a public photo of WPP CEO Mark Read and used it to invite a senior agency leader to a Microsoft Teams meeting. Inside the meeting, the attackers used a voice clone and YouTube footage of another senior WPP executive, and impersonated Read off-camera in the chat window. The targeted leader was asked to set up a "new business" to solicit money and personal details. WPP confirmed the attempt was prevented. Read wrote in an internal email: "We all need to be vigilant to the techniques that go beyond emails to take advantage of virtual meetings, AI and deepfakes." (Slashdot, citing The Guardian email; MM+M Online.)

Ferrari, July 2024 (attempt, unsuccessful). A Ferrari executive received WhatsApp messages from a number with CEO Benedetto Vigna's photo: "Hey, did you hear about the big acquisition we're planning?" A follow-up phone call featured a voice clone with Vigna's southern Italian accent, asking for an "unspecified currency hedge transaction." The executive asked the caller to name the title of a book Vigna had recommended a few days earlier (Decalogue of Complexity by Alberto Felice De Toni). The attacker hung up. (Fortune, 27 July 2024, carrying the Bloomberg report.)

LastPass, April 2024 (attempt, unsuccessful). A LastPass employee received calls, texts, and a WhatsApp voicemail using deepfake audio of CEO Karim Toubba pressing for urgent action. The employee was suspicious because LastPass does not normally use WhatsApp for business communication, and reported it. LastPass intelligence analyst Mike Kosak wrote: "In our case, an employee received a series of calls, texts, and at least one voicemail featuring an audio deepfake from a threat actor impersonating our CEO via WhatsApp." (BleepingComputer, reporting LastPass's statement.)

UAE, January 2020 (US$35 million, successful). The older precedent: a branch manager at a United Arab Emirates company received emails that appeared to come from a company director and from a US-based lawyer coordinating an acquisition, and a phone call in a voice cloned from the director, and authorised US$35 million in transfers to "several bank accounts in other countries in a complex scheme involving at least 17 known and unknown defendants"; Emirati investigators traced two transfers totalling US$415,000 to accounts at Centennial Bank in the United States and asked the US courts for help. (Dark Reading, citing the US court filing.)

Beazley insured, 2023 (US$6 million, successful). A CFO of a Beazley-insured company received a WhatsApp video that appeared to be from the CEO, then continued via chat after "video failed." Over two weeks, the CFO transferred more than US$6 million across multiple payments to a fraudulent Hong Kong account. Beazley described the case publicly on 11 July 2023 and says it paid the claim. (Beazley case study.)

The pattern is consistent: WhatsApp or email setup; an "urgent, confidential" reason; a video call where seeing the boss and other colleagues is the trust-building moment; a "lawyer" or "advisor" lending procedural authority; instructions to wire to specific accounts. The only variable is whether someone in the chain pauses long enough to verify out-of-band.

The attack patterns Singapore SMEs are seeing

Based on the SPF/MAS/CSA joint advisory, the MAS Cyber Risks Associated with Deepfakes paper, and reported cases, the playbook applied to businesses here falls into several recognisable shapes:

  • The "urgent acquisition" pattern. The deepfake CEO calls the finance director or treasurer to authorise a large transfer for a confidential M&A deal, with a "lawyer" sending an NDA. The Arup and Singapore March 2025 cases were close to it: the pretext in the Arup case was a "secret transaction", and in the Singapore US$499,000 case a supposed restructuring of the firm's regional business, with a "lawyer" asking for an NDA.
  • The "supplier change" pattern. A long-standing vendor's "finance director" calls or emails the AP team to confirm new bank account details before the next invoice. Voice deepfaked from a recorded call. This is the structure of the S$57.2 million commodities-firm case (where the trick was a domain look-alike, but the same pattern with voice deepfake is now active).
  • The "regulatory authority" pattern. A deepfake purporting to be a MAS officer, SPF officer, or bank Relationship Manager directs a transfer to a "safety" or "audit" account. Government official impersonation scam losses reached at least S$120 million in the first ten months of 2024 (SPF/MAS, 30 November 2024), S$242.9 million for the whole of 2025, and S$90.8 million in the first half of 2026 - the second-highest loss of any scam type - and the SPF's mid-year brief describes a new opening that impersonates Singtel, YouTrip or Prudential before handing the call to a fake MinLaw, MAS or SPF officer. (SPF Mid-Year Brief 2026; Annual Brief 2025; SPF/MAS, 30 November 2024.)
  • The "internal HR" pattern. A deepfake CEO directs HR or payroll to transfer a "discretionary bonus" or "consultant fee" to a new payee.
  • The "stranded executive" pattern. A deepfake of a director "stuck overseas" needs an emergency transfer for a hotel, lawyer, or visa fixer.
  • The "crypto on-ramp" pattern. The instruction is to convert the funds to USDT or another stablecoin and send to a wallet. Cryptocurrency losses were 24.3% of scam losses in 2024, 20.0% (S$182.2 million) in 2025 and 16.0% (S$65.5 million) in the first half of 2026, with Tether, Ethereum and USD Coin the most-lost tokens. (SPF Mid-Year Brief 2026.)
  • The "subsidiary" pattern. Singapore is a regional HQ for many MNCs. A deepfake of the regional CFO instructs a local Singapore branch to transfer funds to a "subsidiary acquisition." This is functionally the Arup setup.

The SPF, MAS and CSA advise protocols to verify video calls and messages from senior executives through established channels, and CSA says consumer detection tools are "still nascent"; the defence they describe is procedural - out-of-band verification on a pre-established channel - not technical.

Singapore Insurance Market Context

Three insurance product families are involved: Cyber, Crime/Fidelity, and the Social Engineering Fraud (SEF) extensions that bridge the two. None of them automatically pay a deepfake-driven funds-transfer loss. All of them have specific conditions that determine whether they do.

Cyber Insurance. Published Singapore cyber wordings carry breach response, network security failure, business interruption and third-party privacy liability; where a Funds Transfer Fraud or Social Engineering Fraud insuring agreement appears it sits apart, with its own cap inside the main aggregate - AIG's CyberEdge wording, for example, places impersonation fraud and funds transfer fraud under a separate Cyber Crime Coverage section (AIG Singapore CyberEdge wording.) MSIG Singapore's own market commentary acknowledges the dynamic: "As impersonation scams and business email compromise become more sophisticated, insurers may cap limits or narrow the terms for these types of claims. These incidents are difficult to verify and prevent, making them a growing concern for underwriters." (MSIG Singapore, "Ins and Outs of Cyber Policies".) The sub-limit is set in the individual quote, not on the public fact sheet.

The "voluntary parting" problem. This is the single most important wording issue for deepfake claims. A traditional Computer Crime insuring agreement responds only when funds are taken without the insured's authorisation - for example, when a hacker breaches the bank login and transfers funds directly. In a deepfake CEO scam, the insured employee voluntarily authorises the transfer (because they were tricked). Insurers have historically excluded these "voluntary parting" losses unless the policy specifically endorses Social Engineering Fraud or Fraudulent Instruction coverage. (Aon, "When is a cyber crime not a cyber crime?".)

The "verified instruction" condition. Some Cyber and Crime policies condition Social Engineering Fraud coverage on a verification step before the money moves: Aon reports "a few crime insurers" requiring out-of-band authentication, and AIG's Singapore CyberEdge wording makes it a condition that "the Fraudulent Instruction was Verified prior to the Impersonation Fraud Loss", with Verified defined as independent confirmation by a telephone call-back to the requester on a number held on file, in the insured's internal phone directory or verifiable in the public domain, or, for an e-mail instruction, by checking that the requester's genuine work e-mail address was used. (AIG Singapore CyberEdge wording; AIG Singapore CyberEdge product page.) If the AP clerk did not call the CEO back on their real mobile number - or did call back on a number provided by the attacker - the condition is not met and the insurer has grounds to decline the claim.

Crime / Fidelity Insurance. A standalone Commercial Crime policy traditionally covers employee dishonesty, computer fraud, and forgery. The relevant agreements for deepfake-driven losses are:

  • Computer Fraud - loss from a hacker manipulating systems; Aon's reading is that it responds only where the criminal's own unauthorised access effects the transfer without an employee's involvement.
  • Funds Transfer Fraud - fraudulent instructions to a financial institution to move funds from the insured's account; whether a deepfake-induced instruction qualifies turns on the wording.
  • Social Engineering Fraud / Fraudulent Instruction / Impersonation Fraud - written for the scenario where an employee was tricked into transferring funds; offered as an endorsement with its own sub-limit.

Marsh Singapore's commercial crime page puts it this way: "Technology is making fraud easier. Social engineering, deepfakes, and AI-driven tactics are creating more convincing scams, while third-party fraud is becoming more complex," and lists among the risks to review "Coverage gaps: Many organisations assume they are covered, only to find gaps after a loss due to insurer definitions and policy wording differences." (Marsh Singapore Commercial Crime, read 12 September 2026.)

Affirmative deepfake coverage in 2026. One example is Coalition's Deepfake Response Endorsement, announced 9 December 2025. The endorsement amends Coalition's Impersonation Fraud insuring agreement and provides up to US$250,000 for technical analysis by a deepfake forensics firm, legal work to take the deepfake down, and crisis communications support; Coalition frames it as a response to reputational deepfakes - its underwriting chief describes businesses that "ignore fraudulent funds transfer requests" and are still damaged by a deepfake - not as an extension of funds-transfer cover. (Coalition press release, 9 December 2025; IA Magazine, 27 February 2026.) Coalition has explicitly listed the territories where this endorsement is available: the United States, the United Kingdom, Canada (including Quebec), Australia, Germany, Denmark, Sweden, and France. (Coalition press release, 9 December 2025; IA Magazine product profile, 27 February 2026.)

Singapore is not on Coalition's list.

The published Singapore wordings and fact sheets do not brand an endorsement for deepfakes; cover for a deepfake-driven CEO transfer rests on the Fraudulent Instruction, Impersonation Fraud or Social Engineering Fraud insuring agreements within Cyber or Crime policies, subject to (a) verification preconditions, (b) sub-limits below the main aggregate, and (c) claims investigation of whether the loss was a "voluntary parting." Lloyd's Asia lists Beazley among its Singapore service companies, and Beazley has described paying one US$6 million deepfake-related fraudulent-instruction claim. (Lloyd's Asia / Beazley Singapore; Beazley case study.)

The Lloyd's Asia route. Singapore is Lloyd's Market's largest underwriting centre outside London, with 15 syndicates and over 200 expert underwriters. (Lloyd's Singapore.) Lloyd's Asia service companies "are able to trade with any broker and there is no requirement to use a Lloyd's broker", so a licensed Singapore broker can approach a Lloyd's syndicate for higher Crime or SEF limits.

The sub-limit erosion problem. The Arup loss was US$25.6 million. Aon reports Social Engineering Fraud sub-limits in the crime market "starting as low as $10,000 and often having a maximum annual limit of $250,000"; against the Arup loss of US$25.6 million a sub-limit of that order covers under 1 percent. (Aon.) Even on the Singapore March 2025 US$499,000 case, a S$250,000 sub-limit would have left more than half the loss uninsured. This is the primary reason crime/cyber insurance is not a substitute for procedural controls.

Defence costs vs indemnity. Where a policy pays forensic and legal costs inside the same sub-limit as the indemnity, those costs erode what is left for the wired funds; the wording says which.

Notification triggers. Notification deadlines are set in each wording - AIG's Singapore CyberEdge wording requires written notice "as soon as practicable" after a Responsible Officer first becomes aware - so read the notification clause and the definition of awareness, because late notification is a ground for declinature. (AIG Singapore CyberEdge wording.) Notification to the bank comes first: in the March 2025 case the finance director alerted HSBC as soon as the second demand exposed the scam, HSBC notified the Anti-Scam Centre, and the Singapore and Hong Kong centres seized the full amount by 28 March.

Singapore legal and regulatory position

Several Singapore statutes and regulatory instruments touch deepfake-driven fraud:

Penal Code 1871, sections 415, 416, 420. Cheating, cheating by personation, and cheating with delivery of property. A deepfake video instructing a transfer is squarely within section 416 (cheating by personation) read with section 420 (cheating to induce delivery of property). (Penal Code 1871 on Singapore Statutes Online.)

Computer Misuse Act 1993 (CMA). Sections 3 to 7 cover unauthorised access, access with intent to commit an offence, unauthorised modification, unauthorised use or interception, and unauthorised obstruction of a computer; where a deepfake fraud touches the insured's IT systems (for example, compromising a calendar to insert the fake meeting), CMA charges are available. (Singapore Statutes Online, CMA 1993.)

Online Criminal Harms Act 2023 (OCHA). OCHA empowers government agencies to issue Stop Communication Directions, Account Restriction Directions, and Disabling Directions to platforms hosting scam-related content, including deepfakes. The threshold for a scam-related direction is lower than for general criminal harms, recognising the speed required. (CNP Update on OCHA.) On 24 September 2025 the SPF issued the first Implementation Directive under OCHA to a major social media platform, giving it until 30 September 2025 to implement enhanced facial recognition measures in Singapore, prioritise reports from Singapore users and swiftly remove content impersonating Singapore government office holders, on pain of a fine of up to S$1 million plus up to S$100,000 a day; the SPF's Mid-Year Brief 2026 records Implementation Directives to Apple, Google and Meta. (Hogan Lovells, 25 September 2025; SPF Mid-Year Brief 2026.)

Personal Data Protection Act 2012 (PDPA), section 26D. If a deepfake-related compromise exposes personal data - for example, employees' identity documents or customer records used to construct the deepfake or transmitted in the course of the fraud - the organisation must notify the Personal Data Protection Commission "as soon as is practicable, but in any case no later than 3 calendar days after the day the organisation makes that assessment" (s26D(1)), and, where the breach results or is likely to result in significant harm to them, notify each affected individual "in any manner that is reasonable in the circumstances" (s26D(2)), subject to the exceptions in s26D(5) to (7). A breach is notifiable if it results or is likely to result in significant harm to an individual, or affects not fewer than the prescribed number of individuals. The maximum financial penalty under s48J is 10% of annual Singapore turnover for an organisation whose turnover exceeds S$10 million, and S$1 million in any other case. (PDPA Part 6A; PDPA s48J.)

Protection from Scams Act 2025. Came into operation 1 July 2025. Empowers SPF officers to issue Restriction Orders to the seven Domestic Systemically Important Banks (DBS, OCBC, UOB, Citibank, HSBC, Maybank, Standard Chartered) restricting individual banking and credit facilities for up to 30 days at a time (extendable up to 180 days total). This is aimed at retail consumer self-effected transfers, not corporate accounts, but it indicates the direction of travel. (Protection from Scams Act 2025; MHA Press Release, 30 June 2025.) As of 15 February 2026, 12 Restriction Orders had been issued, nine of them still in effect. (MHA written reply, 4 March 2026.)

Shared Responsibility Framework (SRF). Implemented 16 December 2024 via MAS and IMDA Guidelines. Allocates phishing-scam losses between financial institutions, telcos, and consumers when defined duties are breached. The SRF covers consumer retail accounts and phishing scams with a Singapore digital nexus - it does not cover SME corporate accounts in deepfake CEO scams. (MAS Guidelines on SRF; MAS/IMDA Press Release 24 October 2024.)

MAS's AML/CFT Notices to banks, merchant banks and finance companies (Notices 626, 1014 and 824) apply to the institutions, not to you, but they require transaction monitoring, which is why your bank is positioned to notice and freeze a fraudulent transfer, as HSBC's role in the March 2025 case shows. (MAS Notice 626; Notice 1014; Notice 824.)

MAS Information Paper on Cyber Risks Associated with Deepfakes (MAS/TCRS/2025/06, September 2025). Not a binding rule, but it sets out what MAS expects from financial institutions on deepfake risk, including code words, one-time passwords and separation of duties where video- or audio-based instructions are used. (MAS information paper page.)

Concrete scenarios for a Singapore SME

Four short scenarios, reconstructed from the SPF advisories and reported cases:

Scenario 1 - F&B chain, "urgent acquisition." A Singapore F&B chain's Accounts Manager receives a Microsoft Teams calendar invitation marked "Confidential - Acquisition Opportunity," apparently from the founder. On the call are the founder, the CFO, and an "external lawyer." The founder explains the chain is buying a competitor and needs an S$800,000 escrow deposit moved within the hour. The lawyer sends an NDA. The Accounts Manager makes the transfer. All three "people" on the call are deepfaked. Cyber policy with S$250,000 SEF sub-limit, no out-of-band verification performed before transfer - claim partially paid up to sub-limit, S$550,000 retained by the business.

Scenario 2 - Manufacturing SME, "supplier change." A Singapore manufacturing SME's AP clerk receives a phone call from a long-time Vietnamese supplier's "finance director" - the voice is right, the accent is right, the references to last quarter's invoices are accurate. The "finance director" advises that the supplier has changed banks and asks for the next month's payment of S$150,000 to go to a new account in Hong Kong. The clerk updates the vendor master file. The voice was deepfaked from a recording of an earlier phone call; the conversation details came from a previously-phished email account. No out-of-band verification on the pre-established main number. SEF sub-limit insufficient; vendor master data hygiene cited as the gap.

Scenario 3 - Tech SME, "consultant payment." A Singapore tech SME's HR manager gets a WhatsApp video deepfake from the founder, who is "travelling abroad," requesting an urgent payment of S$60,000 to a "consultant" for a confidential pre-Series B project. The founder's WhatsApp profile is correct because the attacker created an account with a public photo. The "founder" provides the consultant's bank details and asks HR to "skip the usual approval" for confidentiality. Below the SEF sub-limit, but loss recovered against the policy only because HR had already conducted dual approval through a code-word system the founder had instituted six months earlier.

Scenario 4 - Logistics SME, "fraud-recovery account." A Singapore logistics SME's treasury team receives a Zoom call from a person identifying as the "DBS Relationship Manager," confirmed by an SMS from a Sender ID that looks genuine. The "RM" warns that the company's account has been used in an external scam and asks for an urgent transfer to a "fraud-recovery holding account" while DBS investigates. This pattern is straight out of the SPF/MAS 30 November 2024 advisory. Because the SRF excludes corporate accounts, the SME has no direct payout claim against the bank under the SRF, and any insurance recovery depends on whether the policy includes Impersonation Fraud and whether out-of-band verification was performed on a number the SME established with DBS before the call.

What This Means for Your Business

A deepfake-driven funds-transfer fraud is not a cyber-defence problem you can fix by buying better firewalls. It is a process and procurement problem, with insurance as a backstop. The actionable shape of the response is consistent across the SPF, MAS, CSA, and the major Singapore brokers:

Step 1 - Out-of-band verification protocols. For any payment over a defined threshold (some companies use S$10,000; others S$50,000; the right number depends on cash flow), require a callback on a pre-established number not provided in the request. Not the number on the calendar invite. Not the number on the WhatsApp profile. The number stored in your HR system from the day the executive was onboarded.

Step 2 - Code-word system. Establish a verbal or written code-word that the executive must provide for any out-of-band payment instruction. Rotate it. The Ferrari case worked because the executive asked for something only the real CEO could know.

Step 3 - Train AP, treasury, finance, and HR specifically. Generic phishing training does not cover deepfake video calls. Training should include real examples of the Arup, WPP, Ferrari, LastPass, and Singapore March 2025 cases and the specific patterns: WhatsApp set-up, "confidential urgent" framing, "lawyer" presence, "new bank account" instruction, "stranded executive" theme.

Step 4 - Banking-side controls. Dual approval on payments above thresholds. Time delays on transfers to new payees, where your bank offers them. Daily payment limit thresholds. Designated "approved payee" list. Money Lock features available on retail-style accounts can be relevant for owner-managed SMEs.

Step 5 - Vendor master data hygiene. Any change to vendor bank account details - any - must be verified by independent contact through pre-established channels, in writing, with a signed letter on the vendor's letterhead, and a phone-call confirmation to a pre-established number. Domain look-alike attacks (the "i"/"l" trick in the S$57.2 million case) succeed because the verification step was skipped.

Step 6 - Tabletop exercises. Run a deepfake fraud simulation at least once a year. Have the CFO send a fake "urgent transfer" instruction to the AP team via WhatsApp from an unknown number; see who calls back, who escalates, who acts on it.

Step 7 - Insurance gap audit. Map the coverage stack for funds-transfer fraud across your Cyber, Crime, and any standalone Social Engineering policies. Establish: where is the FTF coverage; what is the sub-limit; what verification conditions apply; what is the notification deadline; does defence cost share the sub-limit; does the policy respond to a "voluntary parting"; what about supplier-impersonation deepfakes.

Step 8 - Incident response playbook. Step 1 of the playbook is not "call the lawyer" or "notify the insurer." It is "call the bank and ask them to attempt a recall, then file a police report." On the Singapore March 2025 case, HSBC was alerted as soon as the second demand exposed the scam and the full amount was seized in Hong Kong within three days; on the July 2024 US$42.3 million case, the report reached the SPF on 23 July and US$39 million was frozen in Timor-Leste the next day. Insurer notification follows; the wording's notification clock matters, but the bank recall window is measured in hours and days and closes first.

Questions to Ask Your Adviser

When you sit with a licensed adviser or commercial broker to review your policies, ask these questions specifically. Take written answers. The wording shifts year to year and insurer to insurer.

  1. Does my Cyber policy include affirmative Funds Transfer Fraud and Social Engineering Fraud coverage, or is it "silent" on these losses? If silent, will you add an endorsement or recommend a separate Crime policy?
  2. What is the sub-limit for Social Engineering Fraud and Funds Transfer Fraud, and how does that sub-limit compare to my exposure (largest single payment my AP team can make in a day)?
  3. Does the policy explicitly cover losses where my employee was deceived by a deepfake video or voice call - not only by a fraudulent email - and is that coverage aligned with my Cyber, Crime, or both?
  4. What verification conditions does the policy require before a transfer, and what happens to coverage if the employee called back on a number that was provided in the fraudulent instruction itself?
  5. Does the defence cost for investigating the fraud and pursuing recovery sit inside the sub-limit, or is it separate?
  6. What is the notification deadline to the insurer - 24 hours, 48 hours, 72 hours? Who in my company is named as the notification contact?
  7. Does the Cyber policy respond to a supplier-impersonation deepfake (vendor change-of-bank-account fraud), or only to executive impersonation? Where would the line be drawn?
  8. If the loss is partly recovered by SPF/Anti-Scam Centre/INTERPOL after the claim is paid, what is the subrogation position - does the recovery flow back to me, the insurer, or split?

Covarage is a Singapore B2B insurance operations platform. We do not advise on, recommend, or arrange policies. We provide factual information sourced from primary regulators and, where you ask us to, introduce you to a licensed insurance adviser, who gives the advice and places the cover.

Related Information

Published 8 May 2026. Source verified 12 September 2026.