Oracle Financial Services

The Clearwater precedent, and the closest any vendor on this roster has come to escaping it without doing so. Strip every agent and a core banking platform in continuous development since the early 1990s, a full analytical applications suite covering capital adequacy, credit risk and financial reporting, and a financial crime portfolio all continue to run, because they did for three decades.

The vendor's own phrase settles the direction: the capabilities are described as artificial intelligence infused applications, which is intelligence added to an estate rather than an estate built on intelligence. What pushes hardest against that reading, and is recorded because it may change on a later pass, is that the agent catalogue reaches further into the decision than any peer here: a credit decisioning agent shipped in the retail platform in February 2026 and a loan data extraction agent in the corporate platform in April, with hundreds more stated as planned within twelve months. If that catalogue matures into the product rather than an overlay, this grade should be revisited.

A gate is asserted and it is scoped, which is more than the roster norm: the published architecture states that human oversight is embedded for critical decisions in order to meet compliance and ethical governance standards, and separates experience agents that engage clients and bankers from domain agents that collaborate across workflows.

Held at B and not A because nothing defines which decisions are critical, what the default configuration is, what a person must approve, or what an agent may not do. One tension is recorded rather than resolved because the vendor leaves it unresolved: the same material describes autonomous agents that automate mission critical processes and human oversight for critical decisions, without saying where the line between the two falls. On a platform whose shipped catalogue includes a credit decisioning agent, that line is the whole question.

Graded B on the hyperexponential and Provenir precedent, and this is the strongest instance of that pattern the index holds. The vendor ships a dedicated model risk management product as a documented part of its analytical applications estate, which is a full product line rather than a feature or a consulting service, and its published agentic architecture names explainability, policy enforcement, access controls and lineage tracking as embedded through the lifecycle.

Lineage tracking in particular is a concrete auditability mechanism rather than an adjective. Held firmly at B because every part of it faces outward: the governance is over the customer's models, and the vendor publishes nothing about its own. No accuracy figure for the loan data extraction agent that reads hundred page credit contracts, no validation of the credit decisioning agent, no drift monitoring, no versioning and no external assessment of anything in the agent catalogue.

One named customer in the vendor's own material and a very strong independent result, with no number attached to either. NRB Bank in Bangladesh is named in the company's own quarterly customer highlights as selecting the core banking platform at version 14.8, described as the first deployment of that version in the country and as supporting operational efficiency, scalability and service reliability.

Independently, the unit took top rank and category winner placements across fifteen categories of the 2026 Chartis RiskTech100 report. Held at B because no outcome figure attaches to any institution. One point is recorded because it bears directly on the agentic line rather than the platform: trade coverage of both the February retail launch and the April corporate launch noted that no customer adoptions or pilot programmes were disclosed alongside either announcement, so the agent catalogue currently has no published deployment evidence at all. Large third party databases list very senior institutions as users, and those are disregarded here under the standing rule that customer evidence is not graded from an aggregator.

The architecture claims a single governed data layer with stewardship practices embedded at every layer to protect privacy and build trust. Graded C because that sentence describes a design property and binds the vendor to nothing: it does not state whether customer data trains or tunes any agent, whether anything is pooled across institutions on the platform, what an agent retains from a session, or what is excluded from a training corpus.

The standing test applies and the answer is clear enough to state, that the sentence is selling the architecture rather than binding the vendor. The question has real weight for a catalogue of pre built agents, since a pre built credit decisioning agent was built on something and nothing says on what.

Nothing at product level in the material reviewed. The agentic architecture claims data stewardship practices embedded at every layer to protect privacy and build trust, and that is an architectural assertion rather than a data handling position: no retention schedule, no deletion terms, no subprocessor disclosure and no tenant separation statement.

The surface at stake is unusually wide because the same vendor supplies the core banking system of record, the analytics layer and the financial crime monitoring, so the platform holds account, transaction, credit and suspicious activity data for the same institution simultaneously.

No certification, attestation report, audit period or trust portal was located from the financial services product material, so the grade follows the standing rule that a credential must be found and named rather than assumed from a vendor's size.

This is unproven absence rather than evidenced absence and the queued check is both cheap and very likely to move the grade: a cloud and enterprise software provider of this scale certainly maintains compliance documentation, and the reason it did not surface here is that it sits at corporate level rather than on the industry unit's pages. Worth noting as a finding in its own right that a buyer researching the banking platform does not meet a security credential anywhere in the product journey.

A software licensor with no licence, registration or supervised standing of its own. This grade was tested directly against the bank service provider question resolved elsewhere in this sweep and it does not carry across: the United States statutory examination authority attaches to a provider performing services for a depository institution, not to a vendor licensing software the institution runs itself, so it separates processors from licensors.

The open sub question, recorded rather than assumed, is whether the vendor's own hosted and cloud delivery of the core platform constitutes performing the service, which would need testing on the specific arrangement rather than on company size. Everything regulatory in the published material concerns the buyer's obligations, including regulatory reporting, capital adequacy and tax information exchange sold as products.

The phrase ethical governance standards appears in the published architecture and nothing behind it does. No fairness testing, no protected characteristic treatment, no disparate impact analysis, no named framework and no position on any artificial intelligence regulation.

The exposure is the sharpest on this roster and it is specific rather than theoretical: this vendor ships a pre built credit decisioning agent and a collections agent to retail banks, meaning a supplied model can influence who is approved for consumer credit and how a delinquent customer is pursued, across a global installed base including jurisdictions where creditworthiness assessment of natural persons is expressly a high risk use. A vendor that publishes lineage tracking as an architectural principle and publishes no fairness position has built the audit trail and not the audit.

No recourse position published. Two distinct exposures sit under one silence. In retail, a credit decisioning agent and a collections agent reach consumers directly, and nothing states whether a declined applicant learns a model was involved, how a wrong decision is contested or what remedy exists.

In corporate, extraction of terms from hundred page loan contracts into machine readable form carries the wrongly captured instrument risk recorded against Finastra, where the failure mode is a mispriced facility rather than a declined borrower.

Nothing describes how responsibility divides between the vendor supplying pre built agents and the institution deploying them, which is a sharper question for pre built agents than for customer trained models because the vendor, not the bank, chose the behaviour.

No model, provider or version is named for any agent in either the retail or the corporate catalogue, across conversational interfaces, credit decisioning, collections summarisation, compliance automation and contract extraction.

The vendor is unusual in this index in owning cloud infrastructure and model services of its own, so the disclosure question is not only which external supplier is in the chain but whether the agents run on the parent company's own model services or on third party models, and nothing published answers either.

Naming the cloud would not settle it in any case, under the standing rule applied to Opensee, Pennant and SBS, and here the cloud and the vendor are the same company, which makes the omission easier to correct and harder to excuse.

Graded A on the strongest evidence of any vendor on this roster, and the reason is unusual enough to state plainly: the integration is documented publicly at schema level rather than described at category level. The vendor supplies the core banking system of record, the analytical applications layer above it and the financial crime portfolio beside it, and it publishes technical documentation for the integration between the core and the analytics suite naming staging tables, extraction routines, a job control table and its status values.

Every other vendor graded on this roster describes integration in terms of application programming interfaces and system categories. This one publishes the schema a customer would actually implement against. The agentic architecture is separately described as resting on a single governed data layer unifying operational, analytical and event streams.

Deployment flexibility is claimed and residency is not addressed in the material reviewed. The platform is offered in cloud and on premises form with deployment and operational flexibility named as a selling point, and cloud infrastructure is the parent company's own.

Nothing found states region availability, data residency commitments, or how customer data is kept inside a jurisdiction, which matters for a platform sold into markets from South Asia to Latin America to the Gulf with divergent localisation rules. Recorded as a queued check rather than a settled absence, since a cloud provider of this scale publishes region and residency documentation somewhere and it was not located from the financial services pages.

No product pricing published. The commercial model is described by third party review sites as licensing based on deployment scale, modules and user count, and that is an aggregator characterisation rather than vendor disclosure, so it is disregarded under the standing rule.

Nothing published states what the core platform, the analytical packs, the financial crime suite or the new agent catalogue cost, how agents are metered, or whether they are licensed with the platform or separately, which is a live question for a vendor promising hundreds of additional agents inside a year.

Among the widest in the index and stated as an explicit product scope rather than inferred. The core platform is sold across retail, corporate, investment, small business, Islamic banking, microfinance and specialised financial institutions, which is seven distinct institution types with genuinely different product and accounting treatments, and the analytical and financial crime suites extend to insurance as a separate pack.

Functional coverage runs deposits, loans, payments, treasury, trade finance, private wealth, asset management and risk, in multiple currencies and languages, and the customer base is global with named deployments as far apart as South Asia and Latin America. Islamic banking and microfinance sitting in the same platform as investment banking is a real capability distinction rather than a market label.