AI Oversight Gains Urgency as Deepfake Scams Surge 2,000% in Singapore

Deepfake scam volumes have surged more than 2,000% across Singapore in the past three years, and financial institutions are now racing to deploy AI-powered oversight systems capable of detecting threats that traditional risk teams cannot keep pace with. Speaking at the Asian Banking & Finance and Insurance Asia Summit on July 1, Oliver Wyman principal Ashwini Karandikar warned that the next frontier of fraud involves attackers reverse-engineering banks' own models to design attacks that appear legitimate to the very systems meant to catch them.

The threat landscape has expanded beyond simple impersonation. Prompt injection attacks — where hidden instructions are embedded in financial documents invisible to humans but readable by AI — can hijack everything from credit decisions to regulatory filings. Karandikar cited a real-world precedent where Microsoft 365 Copilot disclosed a zero-click vulnerability that allowed a single crafted email to silently exfiltrate internal data, underlining how deeply embedded AI tools create new attack surfaces.

Systemic risk is also mounting in AI-driven trading and hedging, where agents trained on similar data rebalance exposures at machine speed without human hesitation or panic. This allows risk to silently accrue on balance sheets unnoticed. Many banks, Karandikar noted, do not fully know what AI they are running, particularly through third-party vendors deploying tools on their behalf. 'No risk team can review every AI-generated credit decision,' she said, calling for explicit governance frameworks that define where AI can act autonomously and where it must seek approval.

The summit proposed two immediate steps for Singapore's financial sector this year: map every risk touchpoint where AI influences a risk-bearing decision, and design systems explicitly for failure — shifting the governing question from 'Will the system fail?' to 'When it does, how can it fail safely?' Initiatives like Anthropic's Project Glasswing, which uses frontier AI to detect over 10,000 critical vulnerabilities across widely used software, were highlighted as tools that banks should explore alongside their own oversight stacks.

Why it matters for Singapore: As the region's financial hub, Singapore's banks are among the heaviest adopters of AI for customer-facing and risk functions. The same tools driving efficiency and personalisation also create new systemic vulnerabilities. With deepfake scams already costing Singaporeans millions and identity fraud spiking across the region, the call for AI-powered oversight — rather than slower, manual alternatives — signals a maturing recognition that fighting AI-driven fraud requires AI-driven defences. The accountability gap Karandikar identified is not theoretical; it is already shaping regulatory conversations at MAS and across the industry.