The room at Anthropic's invite-only Briefing: Financial Services event in midtown Manhattan was full of bankers. Some had flown in from London. Most had a copy of the agenda in front of them with the speaker line for the day already highlighted. The first conversation was scheduled between Anthropic CEO Dario Amodei and Jamie Dimon, the Chairman and Chief Executive of JPMorgan Chase. They had never appeared on the same stage before.
When Dimon sat down across from Amodei, the room expected a polite back-and-forth about productivity gains and risk frameworks. What it got was an endorsement. "The technology is so powerful, it's worth the trillion-dollar investment," Dimon said. He was speaking about the projected AI capital expenditure across the largest US tech companies for the coming year.
The line did the work Anthropic needed it to do. The head of the largest bank in the country had just told an audience of underwriters, risk officers, and asset managers that the AI buildout was worth it. Then Anthropic showed them what they were buying.
Ten Agents, One Banking Stack
The product launch on Tuesday was unusually concrete. Anthropic released a library of ten pre-built agents designed for the most labor-intensive, repeatable workflows in finance. The names were specific: pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, KYC screener, valuation reviewer, general ledger reconciler, month end closer, and statement auditor.
The list reads like a junior banker's annual To-Do roster. That is the point. Each agent is built around a workflow that today consumes between two and forty hours per week of analyst, associate, or back-office time at a typical investment bank, asset manager, or insurer. Anthropic's pitch is not "use Claude to write better." It is "stop paying humans to do this part."
| Agent | What It Replaces |
|---|---|
| Pitch Builder | Drafting client decks from CRM data and recent filings |
| Meeting Preparer | Compiling pre-meeting briefing books |
| Earnings Reviewer | Reading 10-Qs and writing first-draft summaries |
| Model Builder | Building DCFs and comparable-companies models in Excel |
| Market Researcher | Pulling sector, peer, and macro data into research notes |
| KYC Screener | Reviewing customer onboarding files against PEP and sanctions lists |
| Valuation Reviewer | Running sensitivity tables and audit trails on existing models |
| General Ledger Reconciler | Matching subledger entries against the GL for month-end |
| Month End Closer | Sequencing close tasks, flagging blockers, prepping JEs |
| Statement Auditor | Tying out tickmarks against source documents in 10-K reviews |
Each agent ships in two flavors. The first runs inside Claude Cowork or Claude Code, where a human analyst stays in the loop and the agent assists in real time. The second runs as a Claude Managed Agent, a hosted-by-Anthropic deployment in which the model handles a defined task more autonomously, with Anthropic operating the underlying production infrastructure. The managed-agent track is what was new on Tuesday.
Microsoft 365 Goes From Demo to Default
The second announcement was the one most analysts will be living with. Anthropic shipped general availability of its Microsoft 365 add-ins for Excel, PowerPoint, and Word. Outlook is the next add-in coming. The integration matters because, in a finance workflow, the unit of work is rarely a single application. An earnings model lives in Excel, a client deck lives in PowerPoint, a memo lives in Word, and the supporting email thread lives in Outlook.
What Anthropic shipped is a single Claude agent that works across all four with shared context that persists across applications. Drop a 10-Q into Excel and ask Claude to build a model. Switch to PowerPoint and ask it to draft the deck for the call. The deck inherits the model, the names, the assumptions, and the comparable companies that were already loaded into context, with no copy-paste in between.
In Excel specifically, Claude builds financial models from filings and data feeds, audits formulas across linked workbooks, and runs sensitivity analyses without leaving the cell. In PowerPoint, it drafts decks that update automatically when the underlying numbers change. The integration is meaningful because it cuts directly into the workflow Microsoft itself has been monetizing through Copilot. Anthropic just took the same surface and made the engine swappable.
Moody's, FIS, Dun & Bradstreet, Third Bridge
The third leg was data. Anthropic announced data partnerships that put institutional research, ratings, and reference data directly into Claude as native plug-ins. Moody's is the headline: its full ratings and risk data covering more than 600 million companies is now embedded in Claude as a native app. Dun & Bradstreet added its credit, supply chain, and entity-resolution data. Third Bridge added expert call transcripts.
The most operationally consequential partnership was the one with FIS, the banking infrastructure giant that processes payments, deposits, and core banking workloads for thousands of US banks. FIS announced it is co-building a Financial Crimes AI Agent with Anthropic that compresses anti-money-laundering investigations from days to minutes. Anthropic's Applied AI team is embedded with FIS in a forward-deployed-engineer model. BMO and Amalgamated Bank are listed as the early deployment partners. Broader availability is planned for the second half of 2026.
The combination of named partners is what flipped the tone in the room. Anthropic was not pitching a model. It was pitching a stack: model + Microsoft applications + bank-grade data + bank-grade plumbing.
Mythos, the Other Story Amodei Told That Day
Underneath the financial services launch was a second message Amodei delivered from the same stage. Claude Mythos, the cybersecurity-focused frontier model Anthropic released to a 50-partner list under Project Glasswing in April, has now uncovered tens of thousands of software vulnerabilities. An earlier version of the model found roughly twenty bugs in Firefox. With Mythos, the count climbed to nearly 300, Amodei told the audience. Across all software audited so far, the count runs into the tens of thousands.
Amodei framed the finding as a deadline. "This is about a moment of danger," he said, "where if we respond to it correctly, and I think we started to take the first steps, then we can have a better world on the other side." On the downside, he was direct: "The danger is just some enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that's done from ransomware on schools, hospitals, not to mention banks." The window he gave was six to twelve months before Chinese AI labs reach the same offensive capability. Mythos is currently confined to the Project Glasswing partner list and the federal government. The White House blocked Anthropic from expanding that list to 120 organizations last week. The audience on Tuesday included several Project Glasswing partner companies. The cybersecurity message landed inside the financial services pitch.
The Wall Street Strategy Becomes Visible
Anthropic has been publicly aiming at financial services for a year. What was new on Tuesday was the full picture of the play. The product, the partnerships, and the political endorsement all fit together. Claude Opus 4.7 is the underlying model, released in mid-April with a benchmark profile that beat GPT-5.4 on coding and competitive on financial reasoning tasks. The financial agents sit on top. The Microsoft 365 add-ins make those agents available where bankers actually work. The Moody's, FIS, and S&P data partnerships fill the context window with the references those agents need. Dimon's endorsement gives the entire stack the institutional cover that procurement committees at large banks require.
The strategy lands at a useful moment for Anthropic commercially. The company closed its Series G at $380 billion in February and is reviewing offers to raise another fifty billion at roughly nine hundred billion. Cracking a recurring-revenue beachhead in financial services lifts the multiple Wall Street is willing to assign at the IPO it has scheduled for October. Selling agentic workflows to a JPMorgan-sized customer is materially different, in margin and renewal probability, than selling raw API tokens.
It also gives Anthropic a non-Pentagon enterprise story. The company has been officially blacklisted as a Department of Defense supply chain risk since February, and the DoD signed last week's classified-network agreement with eight AI companies that did not include Anthropic. The financial services play is the explicit alternative customer. Where the Pentagon is unavailable, Wall Street is open.
The Other Side: What the Skeptics Said
Not every reaction in the room was an endorsement. Several CIOs from regional banks told Axios that the same agent library raises the question of who is liable when an autonomous month-end closer mis-classifies a journal entry. Anthropic's answer is the audit trail and human-in-the-loop framing, but the regulatory question is open. The Office of the Comptroller of the Currency has not yet published guidance on autonomous AI agents touching the GL. Regional banks generally do not move ahead of OCC guidance.
A second concern came from inside Anthropic's product line. Claude Mythos can find vulnerabilities at scale, including vulnerabilities in the very banking infrastructure now plugging Claude into its core workflows. Dimon himself flagged the duality in an April CNBC interview: "Mythos reveals a lot more vulnerabilities" was how he framed the offensive side. Tuesday's stage was an explicit reset of that framing into a defenders-have-the-edge narrative. Whether the asymmetry actually holds depends on the patching capacity of every Project Glasswing partner over the six-to-twelve-month window Amodei described.
The third skeptical line came from Microsoft itself. The Microsoft 365 integration brings Claude into the same surface where Microsoft is selling Copilot. Microsoft and Anthropic are nominally cooperating on the Excel, PowerPoint, and Word add-ins. Microsoft is also competing for the same workflow with its own model stack. The arrangement holds for as long as both parties want it to. Microsoft's track record of letting third parties own a core productivity surface is not long.
What Practitioners Should Read Into This
For data scientists, ML engineers, and quants embedded inside banks and insurers, the practical effect is immediate. Anthropic just shipped a managed-agent track that competes for a budget line previously occupied by internal ML engineering or contracted service providers. A risk officer or controller who needed a model-builder agent six months ago is now choosing among three vendors instead of building from a research team.
The deployment shape also matters. Claude Cowork plus Microsoft 365 add-ins is a thin-client install. The IT security review that took six months for an Azure OpenAI deployment is materially shorter for an Office add-in that runs against a vetted enterprise tenant. The path of least resistance into the bank now runs through Office, not through a custom platform team.
Practitioners building internal Claude implementations should also note the Forward Deployed Engineer pattern. Anthropic is shipping its own engineers into FIS to co-build the Financial Crimes Agent. The same model is being offered to large institutional customers through the new Anthropic-Blackstone joint venture announced Monday. The economic message is that integration is now a vendor service, not a customer project.
The Bottom Line
Anthropic walked into Manhattan on Tuesday with a product line, a list of named partners, and the chairman of JPMorgan in tow. By the end of the morning, ten financial agents were generally available, the Microsoft 365 add-ins had cleared GA, Moody's had become a native Claude app, FIS had committed to an AML agent jointly built with Anthropic engineers, and Dimon had publicly underwritten the trillion-dollar AI capex cycle.
The contrast with Anthropic's standing in Washington, where the company is still officially blacklisted by the Department of Defense, was the unspoken subtext of the day. Tuesday's stage was the answer. If the Pentagon does not want to do business with Anthropic, JPMorgan will. If the federal procurement track is closed, the Microsoft Office surface is open. The question for every other AI lab is whether the same playbook still has room for a second occupant.
As Dimon told the room on the consumer-versus-enterprise question: "We always look at it a little bit more like I'm making an investment and what I'm getting for it." The bank just told its peers what the investment is, who is selling it, and what it costs to skip.
Sources
- Agents for financial services and insurance (Anthropic, May 5, 2026)
- Anthropic deepens push into Wall Street with new AI agents, full Microsoft 365 integration, Moody's data partnership (Fortune, May 5, 2026)
- Jamie Dimon blesses the trillion-dollar AI capex boom (Axios, May 5, 2026)
- Anthropic Unveils AI Agents to Field Financial Services Tasks (Bloomberg, May 5, 2026)
- Anthropic CEO warns of cyber 'moment of danger' as AI exposes thousands of vulnerabilities (CNBC, May 5, 2026)
- FIS Brings Agentic AI to Banking with Anthropic, Starting with Financial Crimes (FIS Press Release, May 4, 2026)
- Anthropic deepens its ties to Wall Street with new partnerships, tools (Axios, May 5, 2026)
- Anthropic deepens finance push with 10 new AI agents for banks, insurers (Reuters via Yahoo, May 5, 2026)
- Microsoft and Anthropic team up to bring Claude Cowork to Microsoft 365 (May 5, 2026)
- Jamie Dimon says Anthropic's Mythos reveals 'a lot more vulnerabilities' for cyberattacks (CNBC, April 14, 2026)