Anthropic launches 10 Claude agents for finance — KYC, AML and trading workflows

Banking AI is moving from “help me write a memo” to “go do the work.” That’s the real news here. Anthropic used its May 5 finance event in New York to launch 10 prebuilt Claude agents for financial-services jobs — the kind of work that usually eats analyst, operations, and compliance time. The pitch is simple: stop treating AI like a chatbot and start treating it like a junior teammate wired into your spreadsheets, documents, and internal rules. (anthropic.com) ### What did Anthropic actually launch? It launched 10 “agent templates” for finance and insurance teams. The list is pretty concrete: pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month-end closer, statement auditor, and KYC screener. These are not generic prompts. Anthropic is packaging them as reusable(anthropic.com)nd approval chains. (anthropic.com) ### Why call them agents instead of chatbots? Because the point is not just answering a question once. The point is carrying out a multi-step job. Anthropic describes each template as three parts: skills, connectors, and subagents. Basically, the “skills” hold the workflow and domain rules, the “connectors” pull in governed data, and the “subagents” handle narrower subtasks inside a bigger (anthropic.com)xt generation. (anthropic.com) ### Which finance jobs are they targeting first? The first wave goes after the boring but expensive middle of the stack — pitchbooks, earnings review, financial modeling, month-end close, statement checks, and KYC screening. That mix tells you a lot. Anthropic is not just chasing front-office glamour work like market research. It is also aiming at back-office and control-heavy processes whe(anthropic.com)g documents, and producing audit-ready outputs. (anthropic.com) ### Why is KYC and AML such a big deal? Because that work is repetitive, rules-based, and costly — but mistakes are dangerous. Anthropic’s KYC screener example shows the model assigning a risk rating, checking for missing documents, citing rule outcomes, and recommending whether to clear, escalate, or decline. In other words, Claude is being framed less as a writer and more as a governed de(anthropic.com)ious — banks will still need humans in the loop, because false clears and false escalations both hurt. (theregister.com) ### What else changed besides the agents? Anthropic also expanded Claude’s Microsoft 365 hooks. Claude now works across Excel, PowerPoint, and Word, with Outlook support coming soon. Context can carry across apps, which matters more than it sounds. A finance team can start in a model, pull numbers into a memo, and turn that into a deck without re-explai(theregister.com) data providers finance teams already pay for. (anthropic.com) ### Why is Anthropic pushing finance so hard? Because finance is already a major business for the company. Anthropic said 40% of its top 50 customers are financial institutions, and finance is now its second-largest enterprise revenue segment after tech. That means this is not a speculative side quest. It is a land grab in a vertical that spends heavily on software, data, and labor — and wh(anthropic.com)lue. (money.usnews.com) ### So what’s the bigger market signal? Anthropic is making the same argument it made in coding: once the model is good enough, a lot of standalone software starts looking fragile. Dario Amodei said software development is getting cheaper and hinted that today’s SaaS incumbents may not all survive. Fina(money.usnews.com)s those surfaces, some existing tools risk getting demoted from “system of work” to “data source.” (money.usnews.com) ### Bottom line? This launch is less about one flashy AI feature and more about a change in buying logic. Anthropic wants banks to buy outcomes — faster KYC review, faster close, faster audit prep — not just model access. If that lands, finance software budgets get harder to defend, especially for tools that mainly organize work Claude now claims it can perform. (anthropic.com)