The financial crime landscape has entered a period of unprecedented change. The financial services industry has absorbed successive waves of disruption, from the account takeover surge of the early 2010s to the scam epidemic of recent years. In response, institutions have added signals, tightened controls, and built new detection layers. Several assumptions underlying today’s control fraud and financial crime frameworks are being tested. Authentication assumed a human behind a credential, identity verification assumed a presented document and live face could be trusted when they matched, and dispute processes assumed an authenticated customer either made a transaction or did not. Agentic AI is changing how institutions must approach fraud prevention, anti-money laundering (AML) compliance, and customer protection.
Datos Insights’ ninth annual Financial Crime and Cybersecurity Forum brought together financial crime leaders, solution providers, and industry experts, all focused on how smarter risk intelligence, AI-driven operations, and stronger collaboration can reshape institutional defenses into orchestrated, resilient ecosystems built for the speed of modern threats. Across the Forum, a consistent theme emerged. Responding to the changing threat landscape and increasing pressures from multiple fronts requires stronger data foundations, greater cross-functional collaboration, and continued investment in human expertise alongside technology.
Agentic AI has moved to operating mode: Datos Insights’ recent research found that 88% of institutions are deploying or testing agentic AI tools, yet only 18% can distinguish an authorized agent from malicious automation in real time, and just 25% have dedicated logic for agent-initiated transactions. Only 37% of surveyed institutions have written policies on agentic trust. As institutions cannot allow or restrict agent activity they cannot identify, they must first establish greater visibility into human versus non-human traffic. Further, they will need to determine who bears a loss before the increasing volume of disputes arrive.
Authentication must move from identity to intent: Authorized push payment scams, coached account takeover, and first-party fraud all involve a customer who authenticates correctly, so identity checks alone cannot address them. Defenses have progressed from universal step-up challenges and rigid thresholds toward dynamic controls that weigh context. Signal quality and correlation matter more than signal quantity. Rebuilding authentication around intent and delegated authority becomes an essential component. Define what a customer has authorized an agent to do, treat delegation as bounded rather than blanket, and require proof of mandate as agentic traffic becomes routine.
Verification investments need a pivot: AI has undermined the verification methods the industry has invested in. Because generative AI excels at producing convincing media, controls that evaluate presented documents and images face diminishing returns. For instance, deepfake injection attacks defeat liveness checks and document-to-selfie matching. Institutions should pivot investments to authoritative records that confirm a credential was issued and that a person has a genuine, long-established association with an identifier.
Real-time payments require interdiction, not just detection: Authorized payment fraud is hard to stop because the customer authenticates correctly and instructs the payment willingly. Real-time settlement removes the window in which a mistake can be corrected, with 58% of financial crime leaders describing their instant payment detection capability as inadequate. Closing the gap between detecting risk and interrupting it becomes essential. Fund the interdiction path and the operating model together, resolve off-hours coverage, escalation, and customer contact before enabling send-side capability, and build the integration layer incrementally rather than seeking approval for one large investment.
Information sharing and convergence benefit from structure: Fraud and financial crime leaders recognize the value of information sharing, yet most institutions still operate in isolation. Misconceptions are a larger barrier than technology. Institutions should position information sharing as a growth enabler leaning on reduced friction and expanded payment use cases rather than on avoided losses. Start with a narrow data set, prove value, and then scale. Datos Insights’ research found that 86% of institutions report missed links between fraud and AML functions. Firms must close those gaps through formal forums below executive level, written procedures that specify when each discipline engages the other, cross-training, and shared metrics visible to senior leadership.
Connected intelligence and feedback loops drive improvement: Signals that sit in separate systems frequently leave investigators assembling a picture by hand. The difference between an investigator gathering a picture from five applications and automatically receiving one comes down to careful orchestration of data, systems, and processes. In addition, feedback loops that can carry investigator conclusions back into models help to reduce false positives. Any redesign should begin by defining the target state, separating foundational data gaps that AI cannot address from problems it can solve, and sequencing the work by return on investment.
Data and people determine outcomes: Modernization initiatives fail without data integrity, because layering AI and other advanced technologies over fragmented, incomplete, or bad data leads to poor output rather than improved decisions. Building the data foundation before layering intelligence on top of it becomes vital. Invest in people alongside technology by developing cross-domain expertise and protecting human judgment automated systems cannot replace. Technology can manage volume while people should continue to manage the risk.
Organizational readiness will shape outcomes: Appetite for defensive AI already outpaces results. Technology selection is only part of the decision. Institutions must determine whether to build new capability, acquire it, augment an existing stack, or replace it, and they must define success in measures that are likely to evolve. Although 88% of organizations are pursuing AI for workflow automation, case investigation, and case management, close to 70% cite sustained technology limitations as a major operational challenge. Governance, talent, and organizational readiness will shape results at least as much as the tools chosen.
Financial crime prevention has weathered disruptive shifts before, and each time the industry emerged with a more capable defensive posture. The current compressed timeline changes what preparedness requires. The emphasis is now shifting from acquiring new tools toward building the organizational speed and judgment needed to deploy them responsibly.
- That starts with visibility into human versus non-human traffic, since institutions cannot allow and restrict agent activity they cannot identify.
- It continues with treating consent, liability, and dispute processes as urgent design work, because waiting for regulatory clarity concedes the outcome.
- Governance also does not need to be comprehensive before it is useful.
A serviceable framework built on proven use cases will serve better than a comprehensive one that arrives too late. The practitioners best positioned for the coming period bring their people with them rather than fitting people to technology afterward, and that is the lesson most worth carrying forward.
For a full summary of the discussion highlights, see Datos Insights’ recap of the ninth annual Financial Crime and Cybersecurity Forum. Join us next year for our tenth annual Financial Crime and Cybersecurity Forum from August 31 to September 1, 2027, at the Westin in Uptown Charlotte, North Carolina.
