Anthropic Sends Ten AI Agents Into Wall Street's Back Office

Anthropic on Tuesday rolled out a library of ten ready-made AI agents aimed at the slowest, most paperwork-heavy parts of financial services. The announcement came at the company's first financial services briefing in New York, alongside the debut of its Claude Opus 4.7 model.

The agents target the work that fills analyst pipelines and audit deadlines: pitchbooks and earnings reviews, credit memos, underwriting files, KYC checks, month-end close, statement audits and insurance claims. According to Fortune's reporting, Goldman Sachs, Blackstone and FIS were named as early users.

How the agents work

The Pitch agent can take a target list, build a comparables model in Excel, draft the pitchbook in PowerPoint and write a cover note in Outlook, all without losing track of context as the work moves between formats. The Credit Memo agent edits documents in Word against a bank's own template library, rather than producing generic boilerplate that an associate would then have to rewrite by hand.

Anthropic says the agents ship as plugins inside Claude Cowork and Claude Code, and as cookbooks for its Claude Managed Agents service. The company is pitching this as a way for a team to deploy Claude on live financial work in days rather than the months a typical custom integration takes. American Banker notes that vendor alliances with the major financial-data providers were announced at the same briefing, with Moody's data and Microsoft 365 integration both confirmed.

The bigger play

The release fits a clear pattern. In recent weeks Anthropic has leased the whole of SpaceX's Colossus data centre for compute capacity, opened its models to US government safety testing, and now positioned itself directly inside the regulated workflows of banks and insurers.

That last move is the most consequential. Finance has, until recently, been one of the slowest sectors to adopt frontier AI, partly out of caution and partly because of regulatory constraint. By shipping pre-built agents that map onto specific, named tasks rather than asking firms to design their own from scratch, Anthropic is trying to lower the activation energy for a sector that prefers tools it can audit.

Chief executive Dario Amodei used the briefing to warn that the way software is built and consumed will shift sharply as agents take on more of the labour, a theme that has shadowed every product launch from the major labs this quarter. Whether the agents land with risk officers and compliance teams is a separate question. The next few earnings cycles will be the test.