Anthropic is dramatically expanding its footprint in financial services, launching a suite of pre-built AI agents for the world’s largest banks and debuting Claude Opus 4.7—its most capable model for financial work yet. The announcements, made Tuesday at the company’s invite-only financial services briefing in New York, cap a 48-hour blitz that signals Anthropic isn’t just selling AI software to banks. It’s building the infrastructure, the deployment mechanism, and relationships in the financial industry to become the operating layer for Wall Street.
The strategy has two tracks: one aimed at the largest institutions, giving them tools to configure and run AI agents themselves; the other aimed at the mid-market, using a new private equity-backed joint venture to embed Claude directly into company operations. Together, they represent perhaps the most aggressive push yet by any AI company to capture financial services end-to-end.
The era of consumer-app land grabs is giving way to something more durable for frontier AI labs: enterprise revenue. For both OpenAI and Anthropic, winning paying clients across industries—banks, law firms, software companies, healthcare systems, government agencies—has become the load-bearing pillar of the business model. Enterprise contracts offer what consumer subscriptions cannot: high-margin, multi-year commitments; deep integration into mission-critical workflows that make switching costs real; and usage volumes that justify the staggering capital expenditures these labs are pouring into compute. Anthropic in particular has leaned hard into this positioning, building Claude’s reputation around reliability, safety, and coding performance—qualities that matter far more to a Fortune 500 CIO than to a casual user.
Dimon opened with a personal note. Over the weekend, he said, he had logged onto Claude Code himself. “I want to know about asset swaps and Treasury bid-ask spreads, and quitting the markets, and investment grade.” In 20 minutes, he said, it created a huge dashboard, “with all the backup, and all the research, and it was very accurate about what I wanted.” JPMorgan first began using AI in 2012, he added, and now has use cases ranging from risk and fraud to marketing, design, note-taking, and more.
The product launch filled in what the joint venture will actually be deploying. At the core is a library of roughly 10 pre-built AI agents designed for the most labor-intensive workflows in finance: pitchbooks and earnings analysis, credit memos, underwriting, KYC, month-end close, statement audits, and insurance claims.
Each agent ships as a reference architecture — complete with the skills, connectors, and subagents needed to run that workflow out of the box. Firms can then adapt them to their own modeling conventions, risk policies, and internal approval chains. Once configured, an agent can run as a plugin within Claude’s Cowork and Claude Code environments alongside human analysts, or as a Claude Managed Agent, in which Anthropic handles the secure production infrastructure.
The integration is a significant move for financial services firms, where analysts spend a large portion of their day toggling between spreadsheets, slide decks, and email.
Underpinning all of it is Claude Opus 4.7, which Anthropic says now leads Vals AI’s Finance Agent benchmark with a score of 64.4% and tops the GDPval-AA evaluation for economically valuable knowledge work.
Argenti described three sequential waves of AI deployment at Goldman: first, empowering the technology team (roughly a third of the firm) to operate “at a completely different pace”; second, reimagining operational processes end-to-end; and third—more exciting in the long term, in his opinion—using AI to make better risk and investment decisions. “This is the first time that instead of buying infrastructure, you can actually buy intelligence,” he said. “It goes more profoundly into the way we operate, in the way we think.”
Zaffino disclosed a benchmark that his team ran: Claude out of the box scored 88% as accurate as a human expert on insurance claims. You can look at this two ways: “The theory is, can it get better? Yes. But that assumes that the claims expert doesn’t get better. And so I think it’s also a mechanism for people to learn and for very good questions.”
Beer described JPMorgan’s approach as rewiring the business end-to-end—across how it builds products, serves clients, and develops talent—and said the harder challenge isn’t the technology itself but organizational absorption. “There’s this capability overhang,” she said. “The technology can do so much. It’s the actual organization’s ability to digest and absorb it that tends to be where the gap is.”
On the jobs question, prompted by a recent Wall Street Journal report suggesting AI could trigger 20%–30% unemployment, Dimon and Amodei both expressed uncertainty while rejecting fatalism. “I don’t think anyone knows,” Amodei said. “We need to prepare for the pessimistic case… even start to set policy, because that policy has a lag of two to three years, and the technology is moving so quickly.”
For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.
