Ten. That is how many ready-to-run agents Anthropic shipped when it launched Claude for Financial Services on 5 May 2026, and exactly one of them touches KYC. If you have heard the phrase "Claude's KYC package" and pictured a finished compliance product, the accurate picture is smaller and more specific: a single agent, called the KYC Screener, that parses onboarding documents, runs them against a rules grid, and hands a flagged file to a human reviewer. That is a genuinely useful building block. It is not, on its own, an onboarding platform a Head of Compliance can run a fund through end to end.

This article works through what Claude for Financial Services actually is, what the KYC Screener specifically includes, and how a compliance team would realistically use it, before it needs something built around it.

What Is Claude for Financial Services?

Ten ready-to-run agents, not a compliance platform

Announced by Anthropic on 5 May 2026, Claude for Financial Services is a set of ten ready-to-run agent templates for financial-services work, published as open source at github.com/anthropics/financial-services. Nine of the ten sit on the research, deal, and accounting side of the business: a pitch builder, a meeting preparer, a market researcher, an earnings reviewer, a model builder, a valuation reviewer, a GL reconciler, a month-end closer, and a statement auditor. The tenth, the KYC Screener, sits in the operations vertical plugin and is the only one of the ten built for onboarding and compliance work.

That framing matters for anyone evaluating it as a compliance tool. Claude for Financial Services is not a compliance product with a KYC feature attached to it. It is a broad set of financial-services agent templates, and KYC happens to be one of the ten use cases Anthropic chose to cover, built on a skill Anthropic calls kyc-document-parsing.

Two ways to run it: plugins or the Managed Agents API

Anthropic ships all ten agents from a single source, the public repository, in two delivery formats. A firm can install them as Claude Cowork plugins, or inside Claude Code through the plugin marketplace, with one command: claude plugin marketplace add anthropics/financial-services. Alternatively, a firm that wants to run an agent behind its own workflow engine can deploy it through the Claude Managed Agents API instead, without going through the plugin interface at all.

Anthropic also ships add-ins for Microsoft 365, covering Excel, PowerPoint and Word, with Outlook coming soon, and context carries between the apps. That is most useful for the research and deal-side agents; it is not specific to the KYC Screener.

What the KYC Package Actually Includes

The "KYC package" inside Claude for Financial Services is the KYC Screener agent, delivered as both a Cowork plugin and a Managed Agent, and built on the kyc-document-parsing skill. Per the repository, it does four things.

Parses onboarding documents into an entity file

It reads the documents submitted for onboarding and assembles them into a structured entity file: the raw inputs turned into something a rules engine, and a human reviewer, can both work from, rather than a folder of PDFs.

Runs an automated rules engine

It evaluates the entity file against a rules grid, an automated pass that checks the assembled facts against defined criteria rather than an analyst working through a paper checklist by hand.

Flags gaps and missing information

Where the entity file is incomplete, or a rule fails, it flags the gap rather than proceeding silently or inferring an answer that was not in the documents.

Packages escalations for a human reviewer

It packages the flagged items into an escalation for a person to look at. It does not resolve them itself.

The human-review-only boundary

This is worth stating plainly rather than glossing over, because it is the single most important fact about what this agent is. The repository is explicit: nothing in it constitutes investment, legal, tax or accounting advice, and the agent’s draft analyst work product is for human review only. The KYC Screener organises and flags; it does not make a compliance decision, and Anthropic does not present it as doing so.

That is the correct design for a reference agent, and it is also exactly why it is a starting point rather than a finished platform. Every output still needs a human decision on top of it, and the repository itself does not specify what that review workflow, its escalation path, or its audit record should look like in production. That is left to whoever builds around it.

The surrounding toolkit

Claude for Financial Services also ships a separate financial-analysis plugin with eleven data connectors: Daloopa, Morningstar, S&P Global's Capital IQ, FactSet, Moody's, MT Newswires, Aiera, LSEG, PitchBook and Chronograph, plus Egnyte and Box for document storage, with further partner connectors announced from Dun & Bradstreet, Fiscal AI, Financial Modeling Prep, Guidepoint, IBISWorld, SS&C Intralinks, Third Bridge and Verisk. These sit on the research and deal side of the toolkit rather than feeding the KYC Screener directly, but they show the scale of what Anthropic is building around Claude for financial-services work generally.

The underlying model is not unproven either. Claude Opus 4.7 currently leads Vals AI's Finance Agent benchmark, scoring 64.37%. Financial institutions already using Claude for Financial Services include Citadel, FIS, BNY, Carlyle, Mizuho, Travelers, Walleye Capital and Hg.

How to Use Claude for Financial Services for KYC

Installing the KYC Screener

For a firm that wants to try it, the fastest route is the plugin marketplace: claude plugin marketplace add anthropics/financial-services inside Claude Code, or the equivalent install through Claude Cowork. A firm building its own onboarding workflow engine and wanting the agent as a component within it would instead deploy the KYC Screener through the Claude Managed Agents API.

Running onboarding documents through it

Once installed, the practical use is straightforward: feed it the documents collected for an onboarding case, and it assembles the entity file, runs the rules-grid evaluation, and flags anything incomplete or failing.

Compliance officer review and sign-off

The output lands with a compliance officer, not a decision engine. The reviewer checks the assembled entity file against the source documents, works through the flagged gaps, and makes the actual onboarding decision. That human review is not an optional layer added on top; per Anthropic's own framing, it is the only point at which the work becomes a decision at all.

What happens when it doesn't fit your workflow

The KYC Screener parses and flags a single case at a time. It does not, on its own, run sanctions or PEP screening, monitor adverse media, manage a portfolio of cases through periodic review, or maintain the kind of audit trail a regulator would expect to see after the fact. A firm handling more than a handful of straightforward onboardings quickly needs those pieces built around it, and that is a workflow and governance problem, not a model problem.

Where Steward Fits

Anthropic’s KYC Screener is a genuinely useful building block: document parsing and a rules engine, open source, free to install. It is not an operational onboarding platform, and it was never presented as one. It does not screen for sanctions, PEP or adverse media; it has no ownership or UBO resolution beyond what is in the submitted documents; no case management across an onboarding book; no ongoing or periodic monitoring once a case clears; and no audit trail built to satisfy a regulator's after-the-fact review.

Steward is the applied product layer that turns that kind of capability into a working platform. Where the KYC Screener hands a compliance officer a flagged file, Steward runs the case end to end and hands them a decision to make. Book a demo to see it run on a real ownership structure.