AI fraud tools go mainstream—but humans hold the keys
The gist
AI fraud-fighting tools are going mainstream in finance, but human oversight remains non-negotiable as compliance and audit risks keep people in the loop.
What to know
- Over 80% of financial firms reported AI-related security breaches in the past year, and 54.8% are adding human-in-the-loop controls to govern AI.
- PYMNTS found 42% of issuers cut fraud losses by at least $5 million with AI, but compliance concerns led 86% of firms to delay and 37.9% to cancel AI rollouts.
- New tools like Finastra’s OperatorAssist speed up payment exception handling, but every AI suggestion still requires human approval to stay compliant and auditable.
AI Oversight Becomes Essential
Financial firms are overhauling outdated compliance systems as governed AI shifts from optional upgrade to operational necessity, with human-in-the-loop controls now the linchpin for scaling secure deployment.
By mid-2026, the case for replacing legacy fraud and compliance stacks had shifted from theory to operating necessity. PYMNTS wrote on June 23, 2026, “For years, payment processors competed on uptime, scale and cost,” but the new requirement was decisioning embedded into live payment flows; at the same time, SmartSearch’s 2026 Compliance Report showed why older workflows were breaking down, with only 30% of firms using AI for sanctions screening, 52% struggling with Ultimate Beneficial Owner verification, and UK firms spending an estimated £33.9bn on compliance, including 36% wasted on processes that could be automated.
What changed between July and September was not just more AI deployment, but more governed AI deployment. Fintech Singapore, citing AvePoint on July 7, reported that “more than eight out of ten companies experienced at least one AI-related security breach over the previous 12 months,” that “organizations are moving quickly to deploy AI, many remain unprepared to govern it safely at scale,” and that “Adding human-in-the-loop controls for overseeing AI agents emerged as the most prevalent strategy, cited by 54.8% of respondents”; 69.7% cited compliance concerns, while 86% delayed rollouts and 37.9% canceled plans, making oversight the condition for scaling.
Human Judgment Powers AI Gains
AI slashes fraud losses and automates routine tasks, but only firms that keep humans in the approval loop can handle faster payments, avoid costly errors, and manage rising operational pressures.
The commercial case for AI in payments is no longer theoretical: PYMNTS reported on July 8 that “42% of Issuers Say AI Has Cut Fraud Losses by at Least $5 Million,” a result that matters because payment risk now depends on systems that can “activate that data in real time” and judge transactions before, during and after authorization. That same pressure is about avoiding unnecessary friction as well as stopping fraud, since issuers “falsely declined roughly 15% of legitimate eCommerce transactions, contributing to an estimated $430 billion in lost sales globally each year,” making automated, contextual anomaly detection economically valuable.
The urgency rises further once money moves faster: PYMNTS said the latest Certainty Project found “57% of firms usually detect fraud or payment nonclearance only after settlement,” shifting value toward AI that can surface risk signals upfront while preserving human approval over consequential decisions. The operational case is also intensifying beyond fraud, as “More than 300,000 accountants have left the American profession in five years” and “the work of closing the books is increasingly landing on teams that cannot be staffed,” increasing demand for autonomous support on routine workflows while keeping people in control for compliance and exceptions. That is the same design logic behind Finastra’s Sept. 29, 2026 Sibos 2026 launch of “Repair Recommendations” within AI OperatorAssist to “streamline payments repair across the lifecycle,” aiming to “resolve issues faster” and “manage higher volumes more smoothly,” with every suggested action still subject to review and approval so exception handling stays compliant and auditable.
