Image generated by Deeptech Times using Google Gemini
In the modern financial ecosystem, fraud has become a scattered nuisance and is now a global industrial-scale enterprise. From deepfake-driven scams to cross-border money-mule networks, financial criminals are evolving faster than most institutions can adapt. Yet amid this escalating threat landscape, one technology is beginning to level the playing field: autonomous AI agents.
Pavel Goldman-Kalaydin, head of AI and machine learning at Sumsub, believes these intelligent systems are becoming the new backbone of fraud investigation.
“AI agents are reshaping how financial crime is detected and investigated,” he explains. “They combine automation with contextual decision-making, allowing organisations to analyse identities, monitor transactions, and uncover synthetic networks in real time.”
In today’s high-velocity digital markets, AI agents serve as tireless co-pilots: automating due diligence, scanning millions of data points, and escalating only the riskiest anomalies for human review. Sumsub’s 2024 identity fraud report highlights why this is particularly urgent in APAC, where network fraud is eight times higher than in the Americas or Europe. Seven of the 10 countries most affected by fraud rings are in the region.
“The scale of networked fraud in APAC demands intelligent automation,” Goldman-Kalaydin notes. “These agents are not designed to replace human expertise, but to augment it so teams can focus on complex cases and strategic decisions, while maintaining transparency and compliance.”
Over time, he believes AI agents will become indispensable across onboarding, transaction monitoring and compliance workflows, essentially forming a symbiotic relationship between human oversight and machine speed.
Opportunities and hidden pitfalls
The transition from human-driven investigations to agent-led workflows promises enormous gains in efficiency, scalability and speed. However, Goldman-Kalaydin cautions that the path is lined with risks. “Performance, interoperability and responsibility are the three main challenges,” he says.
The first, performance, concerns reliability under real-world conditions. Even small error rates can cascade in multi-step workflows. “A 10 per cent error per task can lead to a 65 per cent chance something will go wrong across 10 steps,” he explains. Robust testing, ongoing monitoring and human validation are therefore critical.
Interoperability: how well AI integrates with existing systems and regulations, is another weak point. “Garbage in, garbage out still applies,” he says. “Without high-quality data and well-connected systems, even the smartest AI will produce unreliable results.”
Finally, responsibility looms large. As AI takes on more autonomy, transparency becomes non-negotiable. “Black-box decision making erodes trust,” Goldman-Kalaydin warns. “Enterprises need audit-ready outputs and clear human accountability for every AI action.”
Democratisation of crime and fraud-as-a-service economy
Sumsub’s 2025 global fraud index paints a sobering picture. APAC’s fraud exposure is rising sharply: Singapore fell from first to tenth place among leading digital economies within a year, alongside similar declines in Malaysia, Indonesia and Japan. The reason: digital growth has outpaced defensive innovation.
Emerging threats like deepfakes and synthetic identities are driving this new wave of fraud. “Deepfakes are no longer limited to political interference. They’re being used in job scams, impersonation and corporate fraud,” Goldman-Kalaydin says. “Deepfake fraud jumped 194 per cent year-on-year in APAC, with businesses losing an average of US$300,000 per incident.”
Compounding this is the proliferation of “fraud-as-a-service” operations – bot farms and criminal marketplaces that rent out sophisticated attack infrastructure to anyone willing to pay. The result is a democratisation of crime, where even unskilled actors can execute high-impact fraud.
To keep up, organisations must move beyond static rule-based systems. Sumsub advocates multi-layered, full-cycle verification, which combines identity checks, continuous monitoring and adaptive AI defences. As Goldman-Kalaydin puts it, “76 per cent of fraud happens after onboarding. Companies need protection that lasts the entire customer lifecycle.”
Fighting AI with AI
In a world where GenAI can create lifelike forgeries, defenders must meet innovation with innovation.
“At Sumsub, we believe in fighting AI with AI,” he says. The company’s AI-powered deepfake detection solution, embedded in its proprietary liveness technology, authenticates users in seconds while blocking nearly 99.98 per cent of deepfakes on the first try. It creates a 3D face map of users and cross checks it across future transactions and logins.
Goldman-Kalaydin points to graph neural networks as another breakthrough – capable of mapping relationships between synthetic identities, accounts and devices to expose hidden fraud rings. Combined with real-time anomaly detection and behavioural modelling, these technologies form the core of adaptive, scalable defence systems.
What’s overhyped, in his view, are ‘black-box’ models and plug-and-play AI tools that lack explainability or domain-specific design. “Fraud prevention isn’t one-size-fits-all,” he stresses. “Sustainable defences require interoperability, transparency and continuous learning.”
Building a culture of awareness and adoption
Interestingly, Goldman-Kalaydin argues that the biggest gaps in anti-fraud readiness aren’t purely technical. They’re cultural. “Most companies already talk about data governance and explainability,” he says. “But awareness and adoption are where they fall short.”
He points to two underinvested areas: ensuring that the organisation truly understands the evolving nature of fraud, and actually implementing the technologies it buys. “There’s no point having the best AI if no one uses it effectively,” he says. “Fighting fraud is not the job of one team. It’s an organisation-wide responsibility.”
Balancing speed and oversight
AI’s ability to accelerate investigations is unquestionable but unchecked automation introduces risk. Sumsub advocates a modular governance framework built around transparency and accountability.
For performance-related risks, Goldman-Kalaydin recommends stress-testing agents under real-world volumes and maintaining modular workflows with human checkpoints. For interoperability, standardising data formats and investing in integration frameworks helps ensure smooth collaboration across systems. And for responsibility, he emphasises auditability: “Every AI action must be logged and explainable. Compliance teams should be able to reconstruct and justify any outcome.”
This hybrid governance approach, he says, “combines the efficiency of automation with the trust and control of human oversight.”
Collaboration as the ultimate defence
No single entity can tackle financial crime alone. Goldman-Kalaydin believes that cross-sector collaboration between fintechs, banks, regulators and law enforcement is the strongest line of defence. Shared intelligence and standardisation can reveal complex fraud patterns that individual organisations might miss.
Key enablers include data-sharing protocols, harmonised regulations and interoperable compliance platforms. But obstacles remain: fragmented infrastructure, differing privacy laws and uneven enforcement across APAC. Encouragingly, regional initiatives such as FRONTIER+, a joint anti-scam collaboration between Hong Kong, Singapore and other authorities are proving that real-time intelligence sharing works.
Sumsub’s What The Fraud (WTF) Summit aims to accelerate such collaboration. “We built this platform to bring the ecosystem together,” Goldman-Kalaydin says. “Fraud prevention needs shared purpose and open communication.”
Measuring success in an invisible war
Quantifying fraud prevention is notoriously difficult: you can’t count the crimes that didn’t happen. Yet Goldman-Kalaydin offers a pragmatic framework: track missed cases, cross-check layers of defence, and measure how effectively AI detects fraud that humans would have missed. “Multiply that by the losses or fines avoided,” he says. “That’s your real ROI.”
Managing false positives and adversarial actors is part of an ongoing ‘cat-and-mouse’ dynamic. “Attackers evolve, and so do we,” he says. Continuous development and feedback loops, powered by proprietary data, enable Sumsub to adapt to new tactics faster than fraudsters can exploit them.
Looking five years ahead, Goldman-Kalaydin envisions a world where AI agents become the intelligent fabric of financial security. Advances in GenAI, multi-agent collaboration and explainable architectures will allow for faster, more coordinated defences across ecosystems. Meanwhile, regulatory expectations will tighten to demand transparent and auditable AI decision making.
But the vision remains grounded in partnership, not replacement. “AI agents will be our daily co-pilots,” he says. “They’ll manage the complexity of fraud at scale, while humans stay in control of critical judgments. Together, we can build a fraud prevention ecosystem that’s faster, smarter and more resilient.”
Catch Pavel Goldman-Kalaydin at the WTF Summit in Singapore on November 19-20. See agenda here.
