On 5 May 2026, Anthropic released ten agent templates for financial services — KYC screening, AML investigation, financial modelling, month-end close. FIS announced a Financial Crimes agent built on Claude that compresses Anti Money Laundering (AML) investigations from hours to minutes.
Nobody in the room was talking about where the money was going.
Authorised Push Payment fraud (APP fraud) doesn't break through authentication. It waits until after authentication is done, when a legitimate payer submits an instruction to an account that nobody has checked. That moment sits outside every template Anthropic released. It's a gap that more sophisticated payer-side AI won't close, because the problem lives somewhere else entirely.
Ten Agents, One Blind Spot
The release covers two parts.
The front-office track handles research and client work: a pitch builder that runs comparables and drafts pitchbooks, a meeting preparer that assembles counterparty briefs, an earnings reviewer that reads filings and flags thesis-relevant changes, a model builder that creates and maintains financial models from data feeds, and a market researcher that synthesises news, broker research, and sector developments for credit and risk review.
The back-office track handles operations: a valuation reviewer, a general ledger reconciler, a month-end closer, a statement auditor, and a KYC screener. The KYC screener pulls structured fields from entity files, applies the firm's KYC rules against them, screens named parties for sanctions and PEP exposure, and packages whatever needs human attention into a compliance file. The output is structured and auditable.
Alongside the templates, Anthropic expanded its data partner network by adding connectors to Dun & Bradstreet, Moody's, FactSet, S&P Capital IQ, and others, giving the agents governed, real-time access to the market and identity data that financial professionals already rely on.
Deployment runs two ways. As a plugin in Claude Cowork or Microsoft 365, the agent works alongside the analyst, with context carrying automatically between Excel, PowerPoint, and Word. As a Claude Managed Agent, it runs autonomously on a schedule, with credential vaults, scoped permissions, and a full audit log in the Claude Console. The architecture is well-designed for the problems it is built to solve.
Those problems are all on the payer side.
Can AI agents Like Anthropic's Perform Know Your Payee Verification?
The Anthropic templates work on identity, history, and risk signals tied to the person initiating a transaction. That work is complete before a payment instruction is ever submitted. What happens next and whether the destination account belongs to who the payer thinks it does, is outside the scope of the entire release.
This is the Know Your Payee (KYP) layer, not who sent the payment, but who is receiving it. This is also where APP fraud lives. The payer authenticated correctly. The agent cleared them. The payment instruction went to an account controlled by a criminal, because the account details had been manipulated somewhere in the chain like an intercepted invoice, a spoofed supplier communication, a phone call impersonating a known contact.
The two layers sit at different points in the transaction and answer different questions:
Better Payer Intelligence Doesn't Fix a Payee Problem
A more capable payer-side fraud prevention agent doesn't change payee-side exposure. The data doesn't transfer across layers because the questions are structurally different. Payer authentication draws on identity documents, transaction history, and behavioural signals accumulated over time. Payee verification requires something different: real-time access to account ownership records held across thousands of financial institutions, checked at the precise moment a payment instruction is submitted, fast enough not to disrupt the payment flow.
That infrastructure must be built and connected. No foundation model supplies it. The Anthropic agents are a reasoning layer sitting on top of payer data. The payee layer needs its own data, its own network connections, and its own moment in the transaction lifecycle. These are parallel problems, not sequential ones.
Regulators in markets that have understood this distinction have responded accordingly. The UK and Australia’s Confirmation of Payee (CoP) scheme and the EU's Verification of Payee (VoP) requirement under the Instant Payments Regulation both mandate pre-payment account checks specifically because strengthening payer controls proved insufficient on its own. Across the world, payee verification frameworks exist in various forms at the national and scheme level. What is missing is a global unifying layer that is natively designed for the agentic world and interoperable across all payment rails.
The Verification Question Every Payment Workflow Needs to Answer
As agentic payment infrastructure becomes standard, one question determines whether fraud prevention is complete: does your workflow also verify the payee before the payment executes?
An agent that captures payment intent, runs AML checks, clears the payer, and submits the instruction has done significant work. If the destination account is unverified, the fraud risk hasn't been addressed but it's been deferred to the one moment the agent isn't looking.
The iPiD Payment Agent operates at exactly that point. It captures payment intent from invoices, messages, or screenshots, converts unstructured details into structured payment instructions, runs real-time payee verification and name matching using KYP intelligence across 50+ countries, and flags account mismatches or payee risk signals before a payment is created. Where a payee risk is found, it surfaces the event for correction or user confirmation before funds move, not after.
AI agents in financial services need both KYC agents like Anthropic’s new release and KYP agents like iPiD’s.
See how the iPiD Payment Agent verifies the payee before money moves.
