Industry Guide: Fintech
How to Use Claude in Fintech: A Guide for Product, Risk, and Engineering Teams
Ship faster without loosening the controls regulators care about.
This guide is for people at payments companies, neobanks, lenders, and financial software firms who want to put Anthropic's Claude to work on fraud operations, KYC and compliance, customer support, and engineering. It covers what Anthropic has released for financial services and what named fintechs have said in Anthropic's own posts and customer stories.
Knowing how to use Claude in fintech means applying it where a fast-growing financial company feels the strain: summarizing fraud and chargeback cases for analysts, reviewing KYC documents and drafting regtech documentation, answering customer questions with accurate policy language, and writing and reviewing code with Claude Code. Claude speeds up reading, drafting, and building. The company keeps decisions on customers, funds, and compliance with accountable people, inside its model risk, data protection, and audit controls.
Published
Key Takeaways
- The main fintech uses are fraud case summaries, KYC and compliance documentation, customer support drafting, and engineering with Claude Code.
- Anthropic reports that Block's engineers save hours each week with an open-source agent and that Brex automated most expense transactions.
- Anthropic's May 2026 agent templates include a KYC screener that fintech compliance teams can adapt.
- Fraud detection itself usually runs on specialized models. Claude helps investigators and explains outcomes.
- Customer data and money movement require approved deployments, logging, and human sign-off.
Copy and Paste
Prompts You Can Use Today
Fraud case summary for an analyst
You are a fraud operations analyst at a payments company. Summarize the case below for a decision: account history, the flagged transactions with amounts and timestamps exactly as given, device and login signals, the customer's statements, and how the pattern compares with the fraud typologies listed in our playbook. List the evidence for and against fraud separately. Do not make the final decision. Case data and playbook excerpt: [paste].
Evidence for and against in separate lists makes it harder for a fluent summary to lean one way.
KYC document review against policy
Review the onboarding documents below for a small business applicant against our KYC policy excerpt. For each requirement state met, missing, or inconsistent, quoting the document. Flag ownership or address inconsistencies, and draft a short, friendly request to the applicant for anything missing. Do not approve or decline the application. Policy and documents: [paste].
The applicant request doubles as customer communication, so keep it plain and specific.
Regulatory requirement to engineering tickets
Here is a regulatory requirement and our current product flow for it. Translate the requirement into engineering tickets: for each ticket give a title, the exact requirement text it satisfies, acceptance criteria, the audit evidence we need to retain, and open questions for compliance. Do not interpret the rule beyond its text; flag ambiguity instead. Requirement and flow: [paste].
Linking each ticket to exact rule text gives auditors a clean trail.
Customer support reply with policy language
Draft a reply to the customer message below about a delayed transfer. Use only the facts in the account notes and the help center policy excerpt pasted here. Explain what happened and what happens next in plain language, keep it under one hundred fifty words, do not promise a date the policy does not support, and include the reference number exactly as given. Message, notes, and policy: [paste].
Restricting Claude to the policy excerpt stops it from inventing timelines.
Code review for a payments service
Review the pull request diff below for a payments service. Focus on correctness of money handling: rounding and currency precision, idempotency of retries, race conditions on balance updates, error handling that could leave a transfer half-complete, and logging of sensitive data. For each issue give the file and line, why it matters, and a suggested fix. Diff: [paste].
Naming the specific risks for money movement produces a much sharper review than asking for a general one.
From Anthropic
What Anthropic offers for Fintech
Claude for Financial Services
Anthropic's financial services offering, announced in July 2025, combines Claude models, Claude Code, and enterprise access with connectors to data platforms such as Snowflake and Databricks, plus implementation partners. Anthropic says data is not used for training by default.
· Source
Agents for financial services
Ten agent templates released in May 2026, including a KYC screener, general ledger reconciler, and month-end closer, shipped as plugins in Claude Cowork and Claude Code and as cookbooks for Claude Managed Agents.
· Source
In Practice
How Fintech Teams Use Claude
According to Anthropic's published customer story
Block
In Anthropic's October 2025 financial services update, Block's principal data and ML engineer said most of the company's engineers save many hours each week using goose, Block's open-source AI agent for creating SQL queries.
According to Anthropic's published customer story
Brex
Anthropic's customer story on Brex describes AI expense automation that handles most expense transactions and lifts compliance, saving large amounts of time on expenses and accounting.
According to Anthropic's published customer story
Coinbase
In Anthropic's October 2025 financial services update, a Coinbase senior engineering manager said Claude exceeded the company's performance benchmarks and met its security requirements.
According to Anthropic's published customer story
Visa
Visa's president of technology, quoted in Anthropic's October 2025 financial services update, described AI agents as the next evolution of commerce.
Illustrative walkthrough
Clearing a chargeback queue with a person on every decision
- A fraud analyst opens a disputed payment case in the approved workspace.
- Claude summarizes account history, device signals, and the customer's statement, listing evidence for and against fraud.
- The analyst checks two device signals against the source system and notices one timestamp was read in the wrong time zone.
- Claude drafts the internal decision note and a plain customer letter that reveals no detection methods.
- The analyst edits both, records the decision under their own name, and closes the case.
Fintech and Claude: the Published Numbers
| Figure | Context | Source |
|---|---|---|
| 75% of engineers save 8 to 10+ hours weekly | Block, using its open-source agent goose, per Anthropic's October 2025 financial services update | Source |
| 75% of expense transactions automated | Brex AI expense automation, per Anthropic's customer story | Source |
| 94% compliance rate vs 70% industry standard | Brex expense compliance, per Anthropic's customer story | Source |
| 169,000 hours/month saved on expenses and accounting | Brex, per Anthropic's customer story | Source |
Core Workflows
Claude Fintech Use Cases Across the Company
These AI fintech use cases are where AI for fintech companies saves the most time without weakening controls.
Fraud operations
AI fraud detection in fintech usually runs on specialized models. Claude helps analysts by summarizing cases, comparing patterns with typologies, and drafting decisions and customer letters.
KYC and regtech
Regtech AI work covers KYC automation for document review, policy gap analysis, control documentation, and audit evidence, prepared for compliance sign-off.
Customer support
Replies grounded in account notes and policy text, macros that stay accurate, and escalation summaries for specialists.
Engineering
Claude Code for feature work, code review focused on money handling, test generation, and SQL for analytics and reporting.
AI agents for fintech
AI agents for fintech run multi-step workflows such as KYC screening or expense review that prepare a result for a person to approve, with every step logged.
Finance and operations
Reconciliations, month-end close commentary, and variance explanations drafted from ledger data and checked against the system of record.
What Fintechs Report
How Is AI Used in Fintech Today
How is AI used in fintech? The published examples split into two groups: engineering productivity and operations automation. On the engineering side, Anthropic's October 2025 financial services update quotes Block saying most of its engineers save many hours each week with goose, the company's open-source AI agent for creating SQL queries. Coinbase said Claude exceeded its performance benchmarks and met its security requirements.
On the operations side, Anthropic's customer story on Brex describes AI expense automation, with a large majority of expense transactions automated and a compliance rate well above what Brex cites as the industry standard. Visa's president of technology, quoted in the same Anthropic update, framed AI agents as the next evolution of commerce.
Generative AI in fintech is therefore less about one flagship feature and more about many smaller workflows: a support reply grounded in policy, a KYC file checked against requirements, a code review that knows what can go wrong with money.
Fraud and Compliance
Can AI Detect Fraud in Fintech?
Specialized machine learning models detect fraud in real time, and most fintechs already run them. A general assistant like Claude is not a replacement for those models. Its value is downstream: when an alert fires, Claude can assemble the account history, device signals, and customer statements into a case summary, compare the pattern with your typology playbook, and draft the decision note and customer letter for the analyst.
KYC and regtech work follows the same shape. Claude reviews documents against your policy, flags gaps and inconsistencies, and drafts requests to the customer. Anthropic's May 2026 agent templates for financial services include a KYC screener, which a compliance team can adapt as a starting point, with the approval decision staying with a person.
Keep agent workflows inside your control framework: log every step, route every result to a reviewer, and validate performance on closed cases before relying on it.
Choosing Tools
Claude vs ChatGPT for Fintech
The Claude vs ChatGPT for fintech choice is best made with your own cases and code. Run the same fraud summary, KYC review, and code review prompts in each and compare accuracy, handling of missing data, and code quality. Repeat when your workflow or the tools change.
Fintechs usually decide on deployment and governance: API or enterprise plan, data retention, audit logs, and how the tool fits bank partner and regulator expectations. Many fintechs also build on the API directly, which puts their own controls around every call.
Whichever you use, the discipline is the same: ground outputs in your policies and data, keep humans on decisions about customers and funds, and test before scaling.
Getting Started
A First Month of AI in a Fintech Team
Engineering first
Engineering teams usually adopt fastest. Start with code review and tests on non-critical services.
Closed cases for ops
Test fraud and KYC prompts on decided cases and compare with the actual outcome before touching live queues.
Ground support in policy
Give support prompts the exact policy text. Replies that cite policy are safer than replies from memory.
Agree the controls
Decide with risk and compliance which data, which deployment, and which review steps apply.
Limits
Where Fintechs Must Keep People in Charge
Claude can be wrong with confidence: a misread document, an invented policy detail, or code that looks right and mishandles an edge case in currency rounding. Decisions that move money, close accounts, or decline customers need a person, and code touching balances needs human review and tests.
Fintechs often operate under bank partner agreements and several regulators at once. Any AI-assisted process should be documented well enough that a partner bank or examiner can see how it works and who approves what.
Customer financial data is sensitive and regulated. Use approved deployments, minimize data in prompts, keep secrets out of logs, and never use personal accounts for customer information.
Finally, measure before you scale. Pick one queue or one service, record time per case and error rates before and after, and keep a sample of AI-assisted work for periodic quality review. Evidence from your own operations is what convinces a partner bank, an auditor, and your own leadership that the workflow is under control.
Build the Habits Before You Scale
Precise prompts, grounded outputs, and good review habits are what make AI safe in a regulated product. The free Nightschool AI curriculum is hands-on from the first lesson.
Frequently Asked Questions
How is AI used in fintech?
Fintechs use AI for engineering productivity, fraud case review, KYC and compliance documentation, customer support, and finance operations. Anthropic's published examples include Block's engineers saving hours each week with an open-source agent and Brex automating most expense transactions.
Can AI detect fraud in fintech?
Yes, but usually through specialized fraud models rather than a general assistant. Claude's role is helping investigators: assembling case evidence, comparing patterns with typologies, and drafting decisions and customer letters for a person to approve.
What fintech companies use generative AI?
Anthropic's posts and customer stories name several fintechs using Claude, including Block, Brex, Coinbase, and FIS, with Visa quoted on AI agents in commerce. Most fintechs use some form of generative AI in engineering or operations.
Is Claude used by fintech companies?
Yes. Anthropic's October 2025 financial services update quotes Coinbase and Block, and Anthropic's customer story on Brex describes AI expense automation. Anthropic also publishes financial services connectors and agent templates, including a KYC screener.
Do fintechs lead banks in AI adoption?
Fintechs can often move faster because they build on modern technology stacks and carry less legacy code, but large banks also run significant AI programs. The more useful question for your team is which workflows you can govern well, since speed without controls creates regulatory risk.
How do fintechs use AI for KYC and regtech?
They use it to review onboarding documents against policy, flag gaps and inconsistencies, draft requests to customers, and document controls and audit evidence. Decisions to approve or decline customers remain with compliance staff.
Is Claude or ChatGPT better for fintech?
Test both on your own cases, policies, and code, and compare accuracy and handling of uncertainty. Deployment options, data terms, and fit with your bank partners' and regulators' expectations usually decide the choice.
Is it safe to use AI with customer financial data?
Only through deployments your company has approved, with data minimization, logging, and review. Anthropic states that its financial services offering does not use data for training by default. Keep secrets and full account numbers out of prompts wherever possible.
Keep going
Related Guides and Courses
Private Equity & Finance with Claude
Screen CIMs, run diligence, review models, and draft IC memos with Anthropic's Claude, and know when not to trust the output.
Start the courseStrategic Prompting
Context, iteration, and advanced techniques that get output you actually use.
Start the courseRelated reading
Claude in Fintech, by role
Move Fast With the Controls Intact
Hands-on lessons for product, risk, and engineering teams.