Generative AI is being folded into Southeast Asia’s existing scam infrastructure, allowing operators to produce multilingual phishing material, create synthetic identities and conduct deepfake calls with fewer specialised workers, according to a new United Nations assessment.
The UN Office on Drugs and Crime’s July 2026 regional threat assessment says AI has “dramatically reduced” the technical barriers and staffing requirements for sophisticated fraud. Operators can generate convincing phishing content, deploy real-time face and voice manipulation during live calls and target victims across dozens of languages at once.
The significance is not that AI has created a separate category of crime. It is changing the labour economics of an industry already supported by scam compounds, trafficked workers, underground banking, cryptocurrency settlement and specialised technology suppliers.
Tools sold through criminal marketplaces can generate fake profiles, synthetic documents for bypassing know-your-customer checks and real-time face swaps for video calls, the report says. Work that previously required teams of multilingual operators can now be handled by a smaller number of people using automated translation and persona-generation software. Voice-cloning systems can also produce synthetic speech from samples as short as three seconds, according to UNODC.
The report includes evidence of these tools being used as part of broader criminal systems. One investigation into an Indonesian proxy node linked to a large fraud operation found software configured for real-time face swapping during video job interviews, alongside fake websites, cryptocurrency wallets and tools designed to drain USDT from victims.
That combination matters because language had been one of the natural constraints on cross-border fraud. Scams targeting different countries required scripts, recruiters and operators able to communicate convincingly in each market. AI translation and synthetic personas do not remove that requirement entirely, but they lower the cost of entering additional language markets.
UNODC estimates that scam offences caused between $88.3 billion and $114.1 billion in losses during 2025 across East Asia, Southeast Asia, Australia and New Zealand. The figure covers all scams in those markets; it is not an estimate of losses caused by AI-enabled fraud.
The estimate is calculated by dividing officially reported losses by an estimated reporting rate. The underlying data come from police, anti-scam centres and financial regulators, while reporting rates are derived from national victimisation surveys or regional benchmarks where country-level surveys are unavailable.
UNODC describes the resulting range as an informed approximation rather than a precise measurement, noting that countries use inconsistent definitions and classifications for scam losses.
The report also does not calculate what share of cases involved AI, how much the technology has reduced operating costs, or whether AI-enabled scams succeed more often than conventional ones. Its stronger finding is narrower: generative tools are no longer peripheral experiments. They are being integrated into the recruitment, impersonation and identity-verification layers of an established transnational fraud economy.
