This article was written by Eric Duflos, Senior Financial Sector Specialist and Salvador Chang. The original article was published by the CGAP. You can find the article here.
The race between authorities and criminal organizations to harness the power of AI is on. Criminals have already started weaponizing generative AI to launch social-engineered and highly personalized AI-augmented phishing campaigns that combine synthetic identity with deepfakes.
The sophistication of AI-powered scams increases the likelihood that consumers will fall victim to financial fraud and its consequences. Sumsub reported a 180% year-on-year increase in sophisticated fraud from 2024 to 2025, most of which was generated with AI. Advanced attacks rose from roughly 10% of fraud attempts in 2024 to 28% in 2025. INTERPOL warns that AI is becoming a force multiplier for financial fraud, and that AI-enabled operations generate approximately 4.5 times more revenue than scams without identified AI enablement. The use of AI by criminals is likely to increase risks for consumers. But authorities could catch up by deploying their own AI-powered solutions.
Criminal organizations are not the only ones using AI — authorities and financial providers are also using AI for fraud detection. According to the 2026 Global AI in Financial Services Report by the Cambridge Centre for Alternative Finance (CCAF), which surveyed 628 organizations from 151 jurisdictions, fraud detection is the area where regulators expect to see the greatest benefits (63% of the responding authorities). However, 48% of regulatory authorities are still at the “exploring” stage or not engaged with AI at all, compared to 40% of the financial industry, which is already at advanced stages of adoption.
Over the past two years, CGAP has reviewed 120 initiatives that better protect consumers from fraud and identified more than 50 that have demonstrated measurable success. AI powers more than half of these successful solutions, using machine learning (ML), natural language processing (NLP), and neural network AI technologies. We found that most AI solutions integrate analytical AI capabilities, while some include generative AI models to enhance their capacity to interpret and synthesize unstructured data and understand context. These technologies support phishing and scam attack detection capabilities, biometric identification for onboarding and authentication, behavioral biometrics, and transaction pattern systems implemented by authorities, financial providers, and payment networks along the fraud detection chain. A few real-life illustrations of AI-powered solutions from our report can help actors involved in the digital finance ecosystem harness the technology’s power to protect consumers.
Authorities deploy AI-powered social media monitoring and website scanning systems to identify misleading promotions, unlicensed advice, and investment scams. In 2024, the UK’s Financial Conduct Authority (FCA) automatically scanned 480,000 websites daily, blocked 1,600 illegal sites, and issued 2,240 alerts. In Australia, the Australian Securities and Investments Commission (ASIC) focuses on investment scams and “finfluencers,” removing more than 10,000 websites since 2023.
AI-powered filtering tools help mobile network operators (MNOs) detect and block fraudulent communications and phishing attempts. In Vietnam, Viettel AI analyzes call and SMS patterns. In South Korea, KT’s on-device voice-to-text NLP system detects phishing, flags risky conversations, and sends real-time scam warnings to users and their banks.
Financial providers use biometric verification and authentication to reduce identity theft and account takeover fraud. In Nigeria, Wema Bank uses 3D liveness detection and cross-checks facial biometrics against the national ID database, reporting a reduction in SIM-swap and phishing fraud by 89%.
AI also analyzes behavioral biometrics such as typing speed, device interaction patterns, and transaction patterns to detect anomalies or unusual payments in real time and stop scams in progress. In Malaysia, this has helped AmBank report more than RM 22 million (around 5.4 million USD) in potential fraud prevented since 2019. In Brazil, the Digio card network also used behavioral biometrics to report fraud reductions of up to 90%. Beyond individual institutions, Mastercard TRACE maps transaction flows across banks, traces funds in near-real time, flags mule accounts, detects coordinated fraud rings, and alerts banks in the U.K. and the Philippines.
In Singapore, ScamShield Suite, launched by the National Crime Prevention Council and the Singapore Police Force (SPF), combines an on-device ML analytical AI component with large language models (LLMs). LLMs analyze user-submitted content, such as screenshots from social media, to determine whether it is a scam and communicate the result to the user. This information is combined with the SPF blacklist database to block incoming calls and SMS messages. By June 2025, ScamShield had logged 1.27 million checks and nearly 600,000 user reports and verified more than 230,000 suspicious WhatsApp messages and calls.
Visa Protect brings together real-time transaction data from participating banks and payment service providers to generate instant risk scores, and deploys generative AI to identify signals of enumeration attacks in card-not-present transactions. In a UK pilot, Visa reported it identified 54% of fraudulent transactions that banks’ own systems had missed.
AI applications increasingly operate across institutions. Criminal organizations exploit information gaps among banks, MNOs, platforms, and authorities. AI-enabled intelligence-sharing initiatives help close these gaps. By analyzing structured and unstructured data, AI identifies patterns and generates actionable intelligence for ecosystem-wide anti-fraud collaboration.
Malaysia’s National Fraud Portal (NFP), co-developed by Bank Negara Malaysia and PayNet, illustrates this approach. As the intelligence platform of the National Scam Response Centre (NSRC), it uses graph analytics and ML to trace funds across institutions, flag mule accounts, and share actionable intelligence in real time. Bank Negara Malaysia oversees the platform, and PayNet hosts it. Simultaneously, financial institutions freeze flagged accounts, police investigate and issue freezing and seizure orders, and communications authorities and MNOs block fraudulent numbers. The system has reduced investigation times by 70% and increased fund-freezing success rates from 0.5% to 30%.
Financial authorities and the wider anti-fraud ecosystem must keep pace with emerging AI-enabled risks. INTERPOL warns that agentic AI could allow criminal organizations to plan, execute, adapt, and scale fraud campaigns with limited human intervention. However, the same technology could strengthen prevention by enabling authorities and other ecosystem actors to detect and respond to fraud more quickly and autonomously. According to McKinsey, early banking applications have reduced fraud-detection times by up to 30% and false positives by up to 50%. Such gains require appropriate governance and safeguards.
Building AI capabilities is therefore a priority for financial authorities. A recent CGAP paper identifies five priorities: stronger legal foundations, realistic digital transformation, AI risk management frameworks, an adaptive organizational culture, and deeper collaboration. These measures are essential for authorities to keep pace with AI-driven fraud.
Advanced AI should also be integrated into multi-stakeholder collaborative initiatives. Malaysia’s NSRC shows how this can give authorities an edge over criminal networks. Success requires strong governance, integrated reporting and information-sharing protocols, predictive and prescriptive AI, and coordinated intervention procedures across public and private digital ecosystem stakeholders. Together, these capabilities can better protect consumers.
Tailored B2B information solutions. We collect and integrate vital intelligence, empowering your growth strategies and competitive edge. Accelerate your pathway to success.
