Published in 2026. This report examines developments from 2024–2026.
Executive summary
Between 2024 and 2026, Artificial Intelligence in lending and underwriting matured from a marketing label into several distinct, verifiable business models operating under different risk and regulatory profiles. Direct AI lenders such as Upstart and Affirm continued to originate credit using model-driven decisioning tested through a full interest-rate cycle; network lenders such as Pagaya routed approved risk to institutional capital rather than holding it on a partner bank’s balance sheet; document-AI vendors such as Ocrolus supplied the structured data underneath decisions attributed to more visible consumer brands; and bank origination platforms such as Blend and nCino automated workflows for institutions that retain the credit decision under their own charter. This report documents those categories and the companies behind them using only information traceable to an official company source, a regulatory filing, or established independent reporting. It deliberately avoids repeating unverifiable approval-rate percentages, model-performance claims, or market-size projections that circulate widely in industry commentary but cannot be consistently sourced. Readers should come away with a company-level map of where AI is actually embedded in lending today — and which market signals, from leadership transitions to product refocuses, reveal how each model is holding up under scrutiny.
Coverage period
Published in 2026. This report examines developments from 2024 to 2026. It covers disclosed events from January 2024 through the first half of 2026, with a last-reviewed date of 2026-07-10. Where a company’s history predates 2024 — founding dates, earlier funding rounds, or long-running underwriting platforms — that context is included only where necessary to explain a development within the 2024–2026 window. The report will be revisited as material new developments occur; readers relying on it for time-sensitive credit-policy or vendor-evaluation decisions should independently confirm current details against the official sources listed at the end of this report.
Methodology
Research for this report relied on three tiers of sourcing. The first tier is primary company material: official about and leadership pages, product pages, and investor relations disclosures, treated as authoritative for facts a company would have direct incentive to keep accurate, such as founding dates, executive titles, and product positioning. The second tier is regulatory and legal filings, including SEC disclosures, used for corporate events such as leadership changes, rebrands, and divestitures. The third tier is independent reporting from established outlets, used to corroborate dates and context that a company’s own materials might understate. Figures that could only be found in a single vendor-published claim, without independent corroboration, were excluded rather than repeated. This report does not include any original survey data, and it does not attempt to estimate market size, growth rates, or aggregate fraud-loss totals, because reliable, independently verifiable figures for these categories were not available at the time of writing.
Sector overview
AI lending and underwriting during this period did not converge into a single category. Instead, it split along who holds credit risk, who makes the underwriting decision, and who supplies the data feeding that decision. Direct lenders hold both the decision and, at least initially, the resulting exposure, making their model performance visible in their own disclosed loan-loss data. Network lenders make the decision but route risk to outside institutional capital, creating a different commercial dependency on both partner banks and capital markets. Document-AI vendors improve data quality upstream of someone else’s decision without making credit judgments themselves. Origination-software vendors automate bank workflows for decisions made under a chartered institution’s own brand. Legacy digital lenders layered Machine Learning onto balance sheets and customer bases built well before the current generative-AI boom. Across all five patterns, the common thread is operational embedding: banks and fintechs extended systems they had already begun building before 2024, rather than adopting entirely new categories of tools for the first time. Regulatory attention also intensified rather than relaxed, with fair-lending and adverse-action requirements continuing to apply in full to AI-driven credit decisions in the United States and the EU AI Act’s phased rollout beginning to touch creditworthiness-assessment use cases for institutions serving European customers. Fraud and identity infrastructure increasingly sat in the same procurement pipeline as underwriting itself, as lenders treated synthetic-identity prevention, transaction monitoring, and credit decisioning as an integrated pipeline rather than three separate buying decisions. Explainability vendors from the phase 1 editorial database, such as Zest AI, continued to position model documentation and bias testing as product features rather than compliance afterthoughts, reflecting how regulatory scrutiny shaped vendor roadmaps even when specific compliance requirements had not yet fully crystallized in every jurisdiction.
Category breakdown
Within direct lending, Upstart remains the clearest publicly traded example of AI-driven credit decisioning for personal, auto, and home-equity loans distributed through bank and credit union partners, while Affirm underwrites each purchase individually at point of sale rather than issuing a revolving credit line, and Figure extends the direct-lending model into home equity with blockchain-based origination infrastructure. Within network lending, Pagaya integrates via API with partner lenders, uses models trained across its network to identify creditworthy applicants a partner’s own model would have declined, and absorbs resulting credit risk through institutional capital — a structurally different sales motion than a direct lender pitching consumers. Nova Credit plays an adjacent role, translating cross-border and cash-flow data into lender-ready analytics without assuming credit risk. Within Document AI, Ocrolus converts bank statements, pay stubs, and tax documents into structured underwriting data, often sitting invisibly upstream of decisions attributed to more consumer-visible brands. Within bank origination software, Blend and nCino automate origination and decisioning workflows for chartered institutions rather than making credit decisions under their own brand. Within legacy digital lenders, Enova International and Happen Bank (formerly LendingClub) demonstrate that some of the more battle-tested AI underwriting models run inside companies that predate the current AI marketing wave. Adjacent identity and fraud infrastructure — represented by phase 1 profiles such as Socure and Alloy — increasingly sits in the same procurement pipeline as underwriting itself, reflecting a broader pattern of vendors bundling categories that used to be bought separately.
Company examples
Upstart, co-founded by Dave Girouard, Paul Gu, and Anna Counselman in 2012 and public on Nasdaq since December 2020, disclosed a planned leadership transition in February 2026: Girouard moved to Executive Chairman effective May 1, 2026, with Gu — previously Chief Technology Officer — becoming CEO. Affirm, founded by Max Levchin, underwrites point-of-sale installment credit with a revenue model that does not depend on late fees the way a traditional credit card does. Pagaya routes AI-driven approvals to institutional capital while partner banks retain the customer relationship. Ocrolus supplies document-extraction infrastructure named by lenders including Brex, PayPal, Plaid, and SoFi as customers. Blend, after diversifying into title insurance and later divesting Title365 in 2026, refocused around its Autopilot AI agent for borrower-document review. nCino‘s Bank Operating System, used by institutions across multiple countries, introduced AI foundations with nCino IQ in 2018, giving later features such as Banking Advisor a more mature technical base than vendors starting from scratch after 2023. Figure automates home-equity origination end to end. Nova Credit focuses on cross-border and cash-flow underwriting data. Enova International runs its Colossus machine-learning underwriting platform for non-prime consumer and small-business credit. Happen Bank rebranded from LendingClub in 2026 as a chartered digital-first bank. On the fraud-and-identity side, Socure and Alloy from the phase 1 database illustrate how underwriting decisions increasingly depend on adjacent identity verification and orchestration layers procured alongside the credit model itself. Phase 2 additions Trulioo and Mitek Systems extend identity verification for global and document-based onboarding flows, while Sardine bundles fraud, compliance, and underwriting signals into a single agentic platform. Payments infrastructure from phase 1 profiles Stripe and Plaid remains relevant to lending indirectly: Stripe’s Radar fraud screening and Plaid’s account-connectivity data feed both fintech lending products and, as of 2026, broader personal-finance integrations that depend on the same underlying financial data quality.
Key developments
Several disclosed events define this period. First, Upstart’s CEO transition from Dave Girouard to Paul Gu, announced in February 2026 and effective May 1, 2026, marked one of the most closely watched governance changes among publicly traded AI-native lenders, framed by both co-founders as a long-planned handoff rather than a response to any single quarter’s performance; Girouard remained on Upstart’s board as Executive Chairman. Second, the sustained rate-tightening environment of 2022 through 2024 gave AI underwriting models their first genuine multi-year test outside a benign credit environment, a test that public-market investors and bank partners watched when deciding how much lending volume to route through AI-driven decisioning versus traditional scorecards. Third, Blend’s 2026 divestiture of Title365 and refocus on origination software, described publicly as a reset after spreading across too many product lines, signals that even established fintech infrastructure vendors are narrowing scope rather than expanding indefinitely. Fourth, LendingClub’s 2026 rebrand to Happen Bank reflects a chartered-bank identity that differs materially from its original marketplace-lending positioning, illustrating how long-running digital lenders continue to evolve their regulatory and brand posture. Fifth, Pagaya’s network-lending model continued to depend on convincing both partner lenders and institutional capital markets that risk routed through its API is sound — a dependency that becomes more visible when credit conditions tighten. Sixth, document-AI vendors such as Ocrolus continued to offer both human-reviewed and machine-only processing tiers, reflecting a deliberate trade-off between speed and verified accuracy rather than defaulting to full automation from the start. Seventh, phase 2 entrants such as Sardine and Unit21 continued bundling fraud, compliance, and credit signals into agentic risk platforms, reflecting buyer preference for fewer vendor relationships across the lending pipeline. Eighth, cross-border underwriting data vendors such as Nova Credit raised additional capital to expand cash-flow underwriting capabilities, signaling sustained institutional interest in alternative data layers even as direct lenders faced more public scrutiny of model performance through the rate cycle.
Risks and limitations
This report has several limitations worth stating plainly. It relies on company-disclosed and independently reported information rather than direct access to any company’s internal model performance data, meaning claims about how well any AI underwriting model performs in practice cannot be independently verified here and are therefore not asserted. Some companies discussed do not publicly disclose all leadership details, which limits how completely this report can describe their governance. The regulatory environment for AI-driven lending remains unsettled: U.S. banking regulators and the Consumer Financial Protection Bureau continue to apply existing fair-lending and adverse-action requirements to AI underwriting models, and the EU AI Act’s phased rollout is beginning to affect creditworthiness-assessment use cases for institutions serving European customers, meaning some compliance practices described as current in this report may change materially within the report’s own coverage window. This report also cannot assess how individual banks and credit unions are internally weighting AI-driven decisioning against traditional scorecards, since that information is generally treated as proprietary. Network-lending models such as Pagaya’s introduce an additional layer of opacity: a partner bank’s disclosed performance may not fully reflect model behavior routed through a third-party network. Finally, because this report excludes unverifiable statistics, some readers may find it less quantitatively detailed than reports that repeat vendor-published figures; that trade-off is deliberate and is intended to prioritize accuracy over apparent precision.
Conclusion
The period from 2024 to 2026 was not defined by a single transformative AI lending product, but by the steady operational embedding of machine learning across structurally different business models — direct origination, network risk routing, document AI, bank workflow software, and legacy digital lenders — alongside governance events such as Upstart’s CEO transition that tested how founder-led AI-native financial companies handle leadership continuity under public-market scrutiny. Readers should treat company-level developments as more informative than aggregate market claims, and should continue to monitor Upstart, Affirm, Pagaya, Ocrolus, Blend, nCino, and adjacent identity vendors such as Socure and Alloy individually as the sector continues to evolve through the remainder of 2026 and beyond.
Sources
This report draws on official sources including Upstart’s and Affirm’s company and investor relations pages, Pagaya’s and Ocrolus’s official company pages, Blend’s and nCino’s investor relations and about pages, Enova International’s investor relations site, the official Happen Bank rebrand announcement, and Socure’s and Alloy’s official company pages from the phase 1 editorial database, supplemented by independent reporting for corroborating context on the Upstart leadership transition. A complete list of source URLs, with the date each was last reviewed, is maintained in this report’s structured source list and is available to editorial staff for verification.
Disclaimer
This report is provided for general informational purposes only and does not constitute financial, legal, investment, or regulatory advice. It reflects publicly available information as of the last-reviewed date noted above and does not include material non-public information about any company discussed. Company details, leadership roles, and corporate structures can change after publication; readers making decisions based on this content should verify current details directly with the companies or regulatory bodies involved. This report contains no sponsored content and no company discussed paid for or reviewed its inclusion prior to publication.