FINANCE, POLICY & MARKETSPublished by Paul Ivinskas
fc.The Financial CurrentDAILY INTELLIGENCEWhat matters across finance
FIND THE CONTEXT

Search

News, full continuing stories, research text and tables, official policy links, the source directory, organization pages, tools, and glossary definitions. Research matches link to the newest matching version, including retained historical versions.

Includes your Deep dives date, sorting and reading-time filters. Saved searches rerun against the current library when opened.

Saved searches

Loading saved searches…

32 matches for “AI governance” in Deep dives.

Deep dives

  • AI & Tech · Version published 2026-09-30 · 8 min read · estimated

    Alloy: connected onboarding, fraud decisions and the cost of customer friction

    How orchestration, fraud signals and AI assistance affect account opening and ongoing service, with separate evidence for each component.

    Source

    [2] Alloy, Actionable AI product documentation; reviewed September 27, 2026; vendor claims https://www.alloy.com/actionable-ai

  • AI & Tech · Version published 2026-09-30 · 6 min read · estimated

    BioCatch: behavioral intelligence, scam prevention and the customer experience

    How behavioral context may support safer account and payment journeys, with separate evidence for deployed tools, early research and customer friction.

    A behavioral language model is not a chatbot

    …investigative visualization and a newer sequence-model concept may sit under the same AI narrative but have different maturity and validation evidence. Procurement should identify which version and function would actually be deployed, what inputs it requires and which outputs are contractually suppor…

  • AI & Tech · Version published 2026-10-04 · 5 min read · estimated

    ComplyAdvantage Mesh: screening, payment service and investigation workload

    How financial-crime data and case workflows affect onboarding and payments, with separate tests for matching quality, completion time and operating effort.

    Different checks answer different questions

    …The vendor describes both algorithmic matching and rule-based functions, so a single AI label does not identify the work being performed. [1]

  • AI & Tech · Version published 2026-10-04 · 11 min read · estimated

    Cursor and Anysphere: AI coding after the SpaceX acquisition, with financial-sector controls in focus

    Cursor is Anysphere’s AI software-development platform. Following its August 2026 acquisition by SpaceX, the relevant questions include model strategy, enterprise economics, cloud-agent data boundaries and whether faster engineering produces dependable financial software.

    Revision summary

    Cursor is Anysphere’s AI software-development platform. Following its August 2026 acquisition by SpaceX, the relevant questions include model strategy, enterprise economics, cloud-agent data boundaries and whether faster engineering produces dependable financial software.

  • AI & Tech · Version published 2026-10-04 · 10 min read · estimated

    Databricks: the data platform behind enterprise AI, and the controls that matter in finance

    Databricks combines data engineering, analytics, governance and AI on a consumption-based platform. The platform’s substantial adoption does not establish lower risk; data boundaries, operating costs and model dependencies vary by workload.

    Revision summary

    Databricks combines data engineering, analytics, governance and AI on a consumption-based platform. The platform’s substantial adoption does not establish lower risk; data boundaries, operating costs and model dependencies vary by workload.

  • AI & Tech · Version published 2026-09-29 · Historical version · 4 min read · estimated

    Featurespace: behavioral analytics, payment acceptance and the cost of intervention

    Featurespace’s behavioral fraud analytics in a named NatWest deployment, with reported scam metrics, customer commentary and tests that distinguish detection value from realized loss prevention.

    Operating trade-offs

    Real-time risk scoring can reduce exposure but adds latency, integration and model-governance demands. A bank should test performance during peak volumes, vendor outage and degraded data quality; define a conservative fallback and review alert queues. Model outputs should not automatically close alerts…

  • AI & Tech · Version published 2026-10-04 · 5 min read · estimated

    Feedzai: safer payments, legitimate transactions and financial-crime operations

    How a fraud platform fits into payment acceptance and investigations, with distinct measures for detection, customer friction and operating economics.

    Evidence gates · Table row

    Governance · Reason codes, versions, overrides, fairness and monitoring · Challenge, audit and controlled change

  • AI & Tech · Version published 2026-09-30 · Historical version · 6 min read · estimated

    FICO Platform: customer decisions, operating capacity and financial value

    Decision software connects data, models and business rules to customer outcomes. Examine FICO Platform through service completion, peak demand, integration costs and the limits of its reported lending examples.

    Revision note

    Broadened the profile from lending governance to customer journeys and operating economics; added a capacity-versus-cash-savings example, retained dated customer evidence and corrected the historical guidance citation.

  • AI & Tech · Version published 2026-09-30 · 6 min read · estimated

    Fiddler: monitoring AI performance, service quality and costly errors

    AI monitoring can support credit, fraud, trading and customer-service workflows. Evaluate how Fiddler’s documented monitoring and agent controls help detect consequential problems, reduce diagnosis time and improve the service being delivered.

    Coverage, operating ownership and cost

    …tailored to the institution’s risk profile. A bank should map monitoring to its current governance obligations and model inventory, rather than treating a vendor dashboard or reference to older guidance as a certification. The primary attachment excludes generative and agentic AI from its own scope while…

  • Profiles · Version published 2026-10-04 · 8 min read · estimated

    Kobalt Labs: AI inside the third-party risk review

    Kobalt Labs applies AI to the document-heavy work of vendor and fintech-partner oversight. Named customer accounts show implemented uses and reported time savings, while evidence on review accuracy, economics and adoption remains narrower than the company’s broad automation claims.

    The model-risk boundary changed in 2026

    …SR 26-2, superseded SR 11-7 and SR 21-8. It expressly excludes generative and agentic AI from its scope because those technologies are evolving rapidly, while stating that institutions’ broader risk-management and governance practices should guide appropriate controls. The guidance is most relevant to…

  • Profiles · Version published 2026-10-04 · 11 min read · estimated

    Mercury: business banking software and the unfinished transition to a bank

    Mercury links operating accounts, payments, cards and financial workflows through partner institutions. Its proposed national bank adds a new strategic direction, but conditional regulatory approvals are separate from permission to open.

    The customer base has broadened, but the numbers need context

    Management also said 73% of new customers at the end of 2025 came from outside AI or technology startups, and ecommerce represented 21% of new customers acquired during the year. Its claim that one in three U.S. startups used Mercury relied on a defined universe of companies with recent angel through…

  • AI & Tech · Version published 2026-09-30 · 5 min read · estimated

    Microsoft Copilot Studio: agents for financial-service work

    Agent-building tools can support staff research, service preparation and connected workflows. Evaluate Copilot Studio through the task completed, the information available to each user, and the full cost of review and execution.

    Product boundary and evidence

    Microsoft documents Copilot Studio as an agent-building platform with governance, authentication, connector and data-policy controls. Its security documentation describes administrative controls and integration with broader Microsoft governance services. Availability and configuration depend on the tenant…

  • Credit · Version published 2026-09-29 · Historical version · 4 min read · estimated

    Model drift: connecting changing data to financial and customer outcomes

    Input stability, ranking and calibration measure different things; effective monitoring connects them to the lending decision.

    One metric cannot certify a model

    The broader governance point is consistent with NIST’s voluntary AI Risk Management Framework, which emphasizes contextual evaluation and continuing risk management, and the Federal Reserve’s April 17, 2026 revised model-risk guidance. SR 26-2 superseded SR 11-7 and SR 21-8. Neither should be reduced…

  • Policy · Version published 2026-09-30 · 9 min read · estimated

    Model risk across finance: pricing, liquidity, valuations and SR 26-2

    Models shape financial decisions far beyond underwriting. Examine their purpose, sensitivity and real-world use across funding, payments and valuation, then apply the current SR 26-2 framework in proportion to the consequences of error.

    What would change the conclusion

    …correction of consequential weaknesses. A future interagency statement bringing generative AI into scope, revised statutory duties or changed production use would require reassessment. As of this review, the correct baseline is SR 26-2, with broader governance applied explicitly to the components outside its…

  • AI & Tech · Version published 2026-09-30 · 5 min read · estimated

    Nasdaq Verafin: AI research assistance, investigation workflows and operating value

    How copilots and proposed agentic roles differ, and how to evaluate complete investigation effort rather than the speed of generating a summary.

    Source

    Nasdaq: expansion of Verafin Agentic AI Workforce; June 10, 2026 https://ir.nasdaq.com/news-releases/news-release-details/nasdaq-verafin-announces-expansion-its-agentic-ai-workforce

  • AI & Tech · Version published 2026-09-29 · Historical version · 5 min read · estimated

    NICE Actimize SAM: monitoring activity and completing financial-crime work

    A layered AML platform combines rules, segmentation, anomaly detection and predictive scoring; each layer needs a different performance test.

    Evidence that would change the conclusion

    …reduction is beneficial or that any product replaces accountable investigation and bank governance.

  • AI & Tech · Version published 2026-09-29 · Historical version · 7 min read · estimated

    Nova Credit: cash-flow information, customer access and lending economics

    Cash Atlas and NovaScore explained through dated Chase, PayPal and Imprint relationships, with clear boundaries between permissioned data, proprietary scoring and the final lending decision.

    The data contract matters as much as the score

    The Cash Atlas V1 API reference exposes details useful to governance: source identifiers, income-model versions, recent and annual income fields, confidence measures and report-related status information. Its documentation distinguishes three complete months of recent income from a twelve-month annual…

  • AI & Tech · Version published 2026-09-30 · Historical version · 7 min read · estimated

    Oscilar: fraud decisions, investigations and the customer experience

    Risk detection and investigative agents support different parts of a financial service. Evaluate Oscilar through detection quality, investigator productivity, legitimate customer access and the evidence behind each action.

    Revision note

    Expanded the profile beyond alert governance to payment and onboarding service outcomes; added a worked legitimate-customer-friction example and separated faster evidence preparation from completed case resolution.

  • Policy · Version published 2026-10-01 · 6 min read · estimated

    Regulation P: data sharing, customer choices and financial-service design

    Privacy exceptions support everyday financial services, but the purpose of a data transfer determines what the exception can justify.

    Annual notices and changing practices

    …change since the most recent notice. This is not a permanent exemption from privacy governance. A new sharing arrangement can change whether the exception remains available and whether a revised notice is required.

  • Credit · Version published 2026-09-29 · Historical version · 4 min read · estimated

    Reject inference: credit access, missing outcomes and the limits of an approval claim

    A model trained on approved borrowers cannot directly reveal how rejected applicants would have repaid the proposed loan.

    What would change the conclusion

    …reveal with certainty what every rejected borrower would have done. Readers evaluating an AI underwriting claim should ask how much of the claimed improvement was measured and how much was inferred.

  • AI & Tech · Version published 2026-09-27 · Historical version · 5 min read · estimated

    Resistant AI: document authenticity, customer verification and processing economics

    What document-authenticity models can detect, how their verdicts differ from verified income, and how to handle benign edits, forged statements and customer review paths.

    Source

    [1] Resistant AI developer documentation, About Resistant Documents; reviewed September 27, 2026 https://developers.resistant.ai/getting-started/about

  • Credit · Version published 2026-09-30 · 6 min read · estimated

    Roll rates and cures: how payment problems develop and resolve

    How delinquency migration reveals changing customer needs, servicing workload and portfolio performance that an ending ratio can conceal.

    Forecasting requires more than matrix multiplication

    …different risk from an automated forecast driving material reserves or credit decisions. Governance should follow intended use and consequences, not whether the tool has an AI label.

  • AI & Tech · Version published 2026-09-30 · 6 min read · estimated

    Sardine: transaction sequences, payment acceptance and the economics of fraud decisions

    How device context and transaction-history models may improve fraud decisions, with careful interpretation of vendor research and customer-friction costs.

    Source

    [3] Sardine AI Labs, model approach and reported evaluation results; undated page reviewed September 27, 2026 https://www.sardine.ai/ai-labs

  • AI & Tech · Version published 2026-09-30 · 6 min read · estimated

    SAS Model Manager: putting models to work across finance

    Financial models support forecasts, fraud detection, pricing and customer operations. Examine how SAS Model Manager and Model Risk Management can connect development, deployment and business use, and where integration effort still matters.

    Where the integration can help

    …ownership/version information, risk information visible alongside model operations and governance workflows triggered by changes. These capabilities can reduce manual reconciliation if implemented accurately. They do not prove that validation is rigorous or that the correct individual approved the actual…

  • AI & Tech · Version published 2026-10-01 · 16 min read · estimated

    Spring Labs: AI conversation intelligence, service improvement and the economics of adoption

    Spring Labs Holdings’ complaints, quality-assurance and GRC products can turn conversations into operational evidence. Their wider value depends on better service, useful employee feedback and measurable process improvement.

    Compliance automation still requires a defined decision boundary

    …but its accompanying model-risk guidance explicitly excludes generative and agentic AI from its scope. That does not establish an absence of governance needs, and it does not mean that every machine-learning component of a combined system has the same classification. Institutions need to map each component…

  • AI & Tech · Version published 2026-10-04 · 6 min read · estimated

    Stratyfy: decision software, employee capacity and access to finance

    Transparent decision tools can change the work of lending teams and the customer experience. Separate process improvement, access and repayment evidence when evaluating the business case.

    Trade-offs, security and proof

    Transparent features can help analysts identify fragile variables and improve governance, but reducing model complexity may reduce predictive power. The goal is not a simple model at any cost; it is a controlled decision system whose performance, reasons and consumer effects can be tested. Require documentation,…

  • AI & Tech · Version published 2026-09-30 · 8 min read · estimated

    Taktile: onboarding, case work and the economics of automated decisions

    Taktile combines decision workflows, data, case management and agents. Evaluate its proposed uses through completed customer tasks, evidence quality, exception handling and the operating capacity actually released.

    What the product actually does

    …describes a platform combining a decision engine, data orchestration, case management and an AI Agent Manager. Its public materials distinguish low-code rules and model orchestration from generative assistance: an AI copilot helps write or debug logic, while agents perform configured tasks within workflows.…

  • AI & Tech · Version published 2026-10-04 · 5 min read · estimated

    ThetaRay: anomaly detection, payment growth and the economics of investigation

    How unusual-activity detection may support payment and remittance businesses, and why new patterns need context before they become conclusions.

    Economics and governance

    For U.S. bank governance, the Federal Reserve’s April 17, 2026 SR 26-2 superseded SR 11-7 and the earlier BSA/AML model-risk statement, SR 21-8. It emphasizes an approach tailored to the institution’s model risk. Procurement material that still uses SR 11-7 should be mapped to the current guidance rather…

  • AI & Tech · Version published 2026-09-29 · Historical version · 5 min read · estimated

    Unit21: fraud operations, investigation agents and the economics of growing queues

    A risk platform can contain several kinds of AI; banks need to know what each component predicts, writes or changes.

    Revision summary

    A risk platform can contain several kinds of AI; banks need to know what each component predicts, writes or changes.

  • Profiles · Version published 2026-10-04 · 11 min read · estimated

    Upstart: AI underwriting, institutional funding and a bank still under construction

    Upstart combines automated credit decisions with loan distribution, servicing and its own capital. Its improving 2026 earnings sit alongside material balance-sheet exposure, uneven loan-vintage returns and a conditionally approved bank that has not been verified as operational.

    A marketplace with more than one economic engine

    Analysis: the simplest description, an AI lender, leaves out the part of the system most likely to determine whether a loan actually gets made. A credit model can estimate default risk and recommend a price, but a funding provider still needs to accept the expected return. Borrowers must then accept…

  • Profiles · Version published 2026-10-04 · 11 min read · estimated

    Voyager AI: testing the promise of AI-assisted commercial lending

    Voyager AI develops document and workflow software for community lenders. Its $500,000 pre-seed financing and advisory case study establish an early record, while production adoption, customer economics and several assurance claims still require closer verification.

    Source

    [11] Voyager AI governance and human-review commitments; reviewed October 4, 2026 https://voyagercx.ai/ai-governance

  • AI & Tech · Version published 2026-10-04 · 7 min read · estimated

    Zest AI: automated lending decisions and the customer journey

    Model performance is one part of automated lending. Evaluate completed funding, employee workload, customer communication and recurring contribution alongside the published adoption claims.

    Source

    [5] Zest AI: first-half 2026 business update; August 5, 2026; vendor-reported selections https://www.zest.ai/company/announcements/zest-ai-delivers-record-first-half-fueled-by-77-growth-as-lenders-scale-ai/