Published in 2026. This article examines developments from 2024 to 2026.
Introduction
Between 2024 and 2026, AI in financial services shifted from a set of experimental pilots to something closer to embedded infrastructure. The change was less about any single breakthrough model and more about financial institutions and fintech vendors quietly rebuilding underwriting, fraud, and payments systems around Machine Learning that had, in earlier years, often run in parallel to legacy decisioning rather than inside it. This article looks at where that shift is most visible — lending, fraud and identity, and payments — using publicly documented company examples rather than aggregate market forecasts, and closes with what is worth watching through the remainder of 2026.
Underwriting: the AI-decisioning debate matures
Few companies illustrate the maturing of AI-driven credit decisioning better than Upstart, the AI lending marketplace co-founded by Dave Girouard in 2012. Upstart’s model, which layers alternative variables on top of traditional credit data, has now been tested through a full interest-rate cycle — including the sharp rate increases of 2022 and 2023 that pressured loan volumes across the lending industry — giving outside observers a rare, continuously disclosed public record of how an AI underwriting model performs outside of a benign rate environment. In February 2026, Upstart announced a long-planned leadership transition: co-founder and CEO Dave Girouard moved to Executive Chairman, and co-founder and then-Chief Technology Officer Paul Gu became CEO effective May 1, 2026. The company framed the change as a generational handoff between co-founders rather than a response to any single quarter’s results, and Girouard remained on the board in an advisory capacity. The transition matters for the sector because it signals that AI lending has moved past its founder-CEO honeymoon phase into a more conventional public-company governance rhythm, even as its underlying decisioning technology remains its primary differentiator.
Fraud and identity move from detection to prevention
A second area of visible change is the shift from Fraud Detection after the fact toward prevention at the point of onboarding. Socure has continued to expand identity verification aimed at digital account opening, combining Predictive Analytics across digital, social, and offline identity signals to catch synthetic identities before an account is ever funded. Feedzai has pushed further toward consolidation, positioning its RiskOps platform as a way for banks and payment processors to unify fraud prevention, anti-money-laundering monitoring, and case management that had historically lived in separate point tools. Alloy occupies a related but distinct layer, orchestrating multiple identity and fraud vendors into a single configurable decisioning workflow rather than replacing them outright — a reminder that much of the real engineering work in AI fraud prevention is integration and orchestration, not just model accuracy. Together, these companies illustrate a broader pattern: banks and fintechs are increasingly buying platforms that combine several risk signals rather than stitching together single-purpose point solutions themselves.
Payments infrastructure absorbs AI quietly
Stripe and Plaid illustrate a different pattern: AI embedded so deeply into existing infrastructure that it becomes invisible to the businesses relying on it. Stripe’s Radar fraud-detection system has long used machine learning to screen transactions in real time, and the company has continued to extend AI into revenue Optimization and Billing tooling without repositioning itself as an “AI company” in its marketing. Plaid, the financial data connectivity layer co-founded by Zach Perret, took a more visible step into the AI ecosystem in 2026 by partnering with OpenAI to bring Plaid-linked financial account context into ChatGPT-based personal finance tools — a sign that account-linking infrastructure is becoming a building block not just for traditional fintech apps but for general-purpose AI assistants as well. Both companies demonstrate that in payments, AI adoption often looks like incremental feature expansion on top of infrastructure that already handles enormous transaction volume, rather than a rip-and-replace transformation.
Consolidation and ownership signals worth tracking
The 2024–2026 window also included ownership changes that are easy to miss but matter for anyone mapping the AI finance and fraud landscape. Mastercard completed its acquisition of Recorded Future, a threat-intelligence company whose AI-driven analysis increasingly overlaps with financial fraud scoring, in December 2024, explicitly citing plans to fold Recorded Future’s intelligence into Mastercard’s security and identity products. That deal is a useful reminder that fraud intelligence, cybersecurity Threat Intelligence, and payments risk scoring are converging categories rather than fully separate markets. On the other side of the ledger, the U.S. Department of Justice’s move to block Visa’s planned acquisition of Plaid in 2021 remains a relevant precedent shaping how regulators are likely to view further consolidation between card networks and financial data infrastructure providers as AI-driven products blur the lines between payments, data, and identity.
The regulatory backdrop
Regulators have not stood still while AI underwriting and fraud tools have matured. U.S. banking regulators and the Consumer Financial Protection Bureau have continued to emphasize that existing fair-lending and adverse-action notice requirements apply regardless of whether a credit decision is made by a traditional scorecard or a machine-learning model, meaning AI lenders such as Upstart must be able to explain declines in terms regulators and consumers can understand. In Europe, the EU AI Act’s phased implementation has begun to touch financial services use cases classified as higher-risk, including creditworthiness assessment, pushing vendors serving European banks to invest more heavily in model documentation and explainability tooling. Companies such as Zest AI, which focuses specifically on explainable credit underwriting for banks and credit unions, are a useful reference point for how the industry is responding to that regulatory pressure, even though its own public materials do not disclose its founders by name on its primary about page.
Where things stood in 2024
It is worth remembering the 2024 starting point to appreciate how much shifted by 2026. In 2024, most large banks were still running AI fraud and underwriting pilots inside innovation labs, with production decisioning largely governed by legacy scorecards that had been in place for years, and Generative AI inside banks was mostly confined to internal knowledge-search experiments rather than customer-facing or credit-decisioning use cases. Fintech AI lenders such as Upstart were already public but had not yet been tested through a full, multi-year rate-tightening cycle in the way they had been by 2026. Fraud vendors were beginning to talk about “unified platforms,” but most large financial institutions still operated separate procurement relationships for identity verification, transaction fraud, and AML monitoring rather than buying a single integrated RiskOps-style contract. That baseline is the reference point against which the changes described above should be read: less a revolution than a compounding set of decisions that, together, added up to a materially different operating environment by 2026.
Talent and organizational patterns behind the shift
Behind the company-level events described above sits a quieter organizational pattern: banks and fintechs increasingly hired dedicated “AI risk” or “model governance” functions distinct from their traditional credit-risk teams, reflecting the reality that explaining and defending a machine-learning underwriting model to a regulator requires different documentation skills than defending a traditional scorecard. Vendors responded in kind — companies like Zest AI built their commercial pitch specifically around explainability tooling for bank compliance teams, while Alloy’s orchestration layer exists in large part because banks wanted a way to swap or add fraud and identity vendors without re-litigating a full model-risk review each time. This organizational shift, more than any single model improvement, is arguably the most durable change of the 2024–2026 period, because it changes how quickly financial institutions can adopt the next generation of AI tools, not just how well today’s tools perform.
What to watch through the rest of 2026
Three threads are worth following for the remainder of 2026. First, whether Upstart’s new CEO, Paul Gu, shifts the company’s public communication style now that he — rather than co-founder Dave Girouard — is the face of its AI underwriting story. Second, how quickly banks move fraud and identity spending toward consolidated platforms like Feedzai’s RiskOps and Alloy’s orchestration layer, versus continuing to run point solutions from vendors like Socure alongside them. Third, whether further large acquisitions follow the pattern set by Mastercard’s purchase of Recorded Future, in which a payments network buys AI-driven threat intelligence outright rather than licensing it, a move that could accelerate consolidation between the fraud, identity, and cybersecurity categories. A fourth, quieter thread worth tracking is how bank examiners respond to AI underwriting models that have now operated across a genuine rate-tightening cycle, since supervisory feedback from this period is likely to shape model-documentation expectations for the rest of the decade.
Conclusion
The two years from 2024 to 2026 did not produce a single dramatic transformation of AI in finance; instead, they produced a steady embedding of machine learning into underwriting, fraud prevention, and payments infrastructure, alongside a parallel maturation of the regulatory expectations and internal governance functions attached to it. Upstart’s leadership transition, Plaid’s move into AI assistant integrations, and Mastercard’s acquisition of Recorded Future are each, individually, modest news items — but together they describe a sector where AI has stopped being a feature and started being the operating layer. For readers tracking this space, company-level developments like these remain a more reliable signal than aggregate market-size claims, and each of the companies discussed here — Stripe, Plaid, Upstart, Socure, Feedzai, and Alloy — is worth tracking individually as the next two years unfold.