Anthropic released 10 ready-to-deploy Claude agents for financial services and insurance on May 5, 2026, with Citadel, BNY, Carlyle, JPMorgan, Goldman Sachs, Citi, AIG, and Visa already running them in production. Claude Opus 4.7 powering the agents scored 64.37% on Vals AI's Finance Agent benchmark, and the launch ships with full Microsoft 365 integration, a native Moody's app covering 600M+ companies, and connectors for LSEG, S&P, Morningstar, and PitchBook.
TL;DR — what shipped on May 5, 2026
- 10 prebuilt Claude finance agents covering pitchbooks, valuations, earnings analysis, KYC, and month-end close — each a reference architecture combining skills, connectors, and subagents.
- Microsoft 365 add-ins live for Excel, PowerPoint, and Word; Outlook in beta. A single agent context carries across all four applications without re-prompting.
- Opus 4.7 hits 64.37% on Vals AI Finance Agent bench — Anthropic positions this as industry-leading; even human analysts score below 70% on the same benchmark.
- Production deployments confirmed at Citadel, BNY, Carlyle, JPMorgan, Goldman Sachs, Citi, AIG, Visa, Mizuho, Travelers, Walleye Capital, and Hg.
- Moody's data partnership embeds 600M+ company credit-rating and risk records natively into Claude. LSEG, S&P Capital IQ, Morningstar, and PitchBook ship as MCP connectors.
- FIS partnership targets AML/anti-money-laundering investigation workflows specifically.
What Anthropic actually shipped
Anthropic's May 5 announcement is the most aggressive enterprise rollout the company has executed to date. Until this launch, Claude in finance was a horizontal text model — capable, but undifferentiated against OpenAI and Google for Wall Street workflows. With ten prebuilt agents, full Microsoft 365 add-ins, and a Moody's data app, Anthropic has shifted the question from "is Claude good at finance?" to "which workflow do you want to plug in first?" The list of named adopters is the proof point: Citadel and BNY are not deploying experimental tools.
The agents ship via the new Claude financial-services marketplace, with two deployment topologies. The first is plugin mode — agents run inside Claude Cowork (Anthropic's collaboration product) or Claude Code, with a human analyst in the loop reviewing each step. The second is Claude Managed Agents, an Anthropic-hosted production runtime in public beta where agents execute more autonomously against governed connectors. Both paths use the same underlying agent definitions; the difference is who owns the runtime infrastructure.
The 10 finance agents — what each one does
| Agent | What it does | Primary user |
|---|---|---|
| Pitch builder | Generates target lists, runs comparable-company analysis, drafts full pitchbooks with charts, footnotes, and source attribution. | Investment banking junior bankers, M&A advisory |
| Meeting preparer | Assembles client and counterparty briefing books — recent news, earnings, holdings, prior-meeting notes — into a single dossier. | Coverage bankers, sales-side relationship managers |
| Earnings reviewer | Reads transcripts and 10-Qs, updates DCF and comparable-company models, flags guidance changes against street consensus. | Equity research, buyside analysts |
| Model builder | Constructs financial models from filings, transcripts, internal data — three-statement, LBO, merger, and operating models. | Investment banking, private equity associates |
| Market researcher | Tracks sector developments, synthesizes research from internal and external sources into briefing notes. | Strategists, portfolio managers |
| Valuation reviewer | Checks valuation outputs against documented standards (Big Four playbooks, fairness-opinion templates). | Valuation desks, audit firms, fairness advisors |
| General ledger reconciler | Reconciles GL accounts, runs NAV calculations for fund administration, flags variance against expected ranges. | Fund administrators, fund accounting teams |
| Month-end closer | Runs the close checklist, prepares journal entries, drafts variance commentary, packages binder for review. | Corporate finance, controllership |
| Statement auditor | Reviews financial statements for consistency, audit-readiness, and standard-disclosure compliance. | Internal audit, external audit firms |
| KYC screener | Assembles entity files, screens against sanctions and adverse-media databases, packages escalations for human review. | Compliance, AML investigations, onboarding teams |
Each agent is shipped as a reference architecture — three components defined together: skills (markdown files describing the workflow logic and domain knowledge), connectors (governed access to external systems like Bloomberg, Moody's, internal data lakes, S3 buckets), and subagents (specialized Claude calls invoked by the parent agent for narrow sub-tasks like "extract guidance from this transcript paragraph"). The architecture is intentionally inspectable: every skill is a file an analyst can read, modify, and version-control.
Why this matters for fintech and Wall Street
Three things make this launch structurally different from prior frontier-model finance pushes.
First: the named-customer list is unprecedented. Anthropic disclosed Citadel, BNY, Carlyle, JPMorgan, Goldman Sachs, Citi, AIG, Visa, Mizuho, Travelers, Walleye Capital, and Hg as production deployers. Citadel running an LLM in production is a credibility seal — Ken Griffin's firm runs the most paranoid security and compliance regime on Wall Street. JPMorgan's Jamie Dimon went on the record about Claude Code: "In 20 minutes, it created a huge dashboard, with all the backup, and all the research, and it was very accurate." When the JPMorgan CEO publicly endorses an AI vendor by name, IT-procurement gates fall everywhere from Goldman to the regional broker-dealers.
Second: the Moody's app changes the data-availability calculus. Embedding the full Moody's analytics platform — credit ratings, financials, and risk data on 600M+ private and public companies — into Claude as a native MCP application means an analyst can ask "compare the credit profile of these three private targets" without ever leaving the Claude window. Historically, that workflow required licensed access to Moody's, Capital IQ, and a data engineer to plumb them together. Anthropic just collapsed two engineering teams into one prompt.
Third: the Vals AI benchmark score is inflection-grade. Vals AI's Finance Agent benchmark tests end-to-end agent behavior on multi-step finance workflows — pitchbook construction, valuation review, earnings model updates. Opus 4.7's 64.37% score is the first time a frontier model has crossed into the "matches a third-year analyst on most tasks" zone. To put it in perspective, a human analyst with two years' experience scores in the 60-70% band on the same bench. The frontier just walked into a major league.
The Microsoft 365 plugin angle
The Microsoft 365 integration is the most underreported piece of the launch. Anthropic shipped Claude as a single, context-preserving agent across Excel, PowerPoint, Word, and Outlook (the first three generally available, Outlook in beta). The differentiator is not "Claude can edit a spreadsheet" — Copilot does that. The differentiator is that a single Claude session keeps full context as the analyst moves between Excel, PowerPoint, and Word. Build a model in Excel, ask Claude to draft the pitch slides in PowerPoint based on that model, then have it write the cover memo in Word — all within one continuous reasoning context, no re-prompting, no re-uploading.
This is the same architectural pattern Microsoft has been promising with Copilot since 2023, except executed by Microsoft's largest external AI partner (after OpenAI). The implication for Microsoft's own AI roadmap is uncomfortable: Anthropic, not Copilot, is now the agent layer of choice for the most security-conscious finance customers running M365.
Connectors: the data partnerships that came with the launch
- Moody's. Native Claude app spanning credit ratings and risk data on 600M+ companies. Available immediately at GA.
- FIS. Anti-money-laundering investigation workflow integration — the first banking-core software vendor to ship a Claude-native MCP connector.
- LSEG (Refinitiv). Market data, financial filings, and Workspace integration via MCP.
- S&P Capital IQ. Public-company financials, transcripts, deal data.
- Morningstar. Mutual fund and ETF data, manager research.
- PitchBook. Private market data — venture capital, private equity, M&A.
How this lands for OpenAI and Google
OpenAI's enterprise finance push has been Codex-led and aimed at quant infrastructure — Goldman's mass-rollout of GPT-5.5 to its investment-banking division was announced in early 2026, but did not include named agent templates or a Moody's-tier data partnership at launch. Google's Gemini Enterprise has been positioned around Workspace, not Microsoft 365, and Google's finance-vertical agents (announced at Cloud Next 2026) have shipped as preview-only as of May. Anthropic's May 5 launch puts it at least one quarter ahead of either competitor on enterprise finance maturity, especially if the Vals AI benchmark gap holds.
Anthropic CEO Dario Amodei disclosed during the launch that the company hit approximately 80x annualized revenue growth in a single recent quarter, dramatically exceeding internal projections. The finance push is one of the demand drivers behind that curve.
What it costs and how to get it
Anthropic has not published a flat list price for the agent pack. Three deployment paths exist as of May 6, 2026:
- Claude Cowork or Claude Code subscription. Includes the agent templates as plugins. Pricing per seat per month, scaling with seat count and connector usage.
- Claude Managed Agents (public beta). Anthropic-hosted runtime, charged on agent-execution metering. Best for autonomous workflows that fire on a schedule.
- Claude API + open-source agent definitions. The agent skills are reference architectures Anthropic has indicated will be partly open-sourced; teams that want full self-hosted control can wire the same skills against their own Claude API endpoint and connector stack.
What to watch next
- Q2 2026 enterprise-deal disclosures. Anthropic's next funding milestone reporting will likely break out finance-vertical ARR for the first time. Watch for whether it crosses $200M run-rate by Q3.
- Open-sourcing of the agent skills. Anthropic has hinted that core agent definitions will be released under a permissive license, similar to Claude Skills. Timing TBD.
- Compliance attestations. SOC 2 Type II, ISO 27001, FINRA-aligned audit trails, and EU AI Act high-risk classification will determine which agents EU institutions can deploy in production. Expect formal attestations within Q3.
- OpenAI's response. A finance-vertical OpenAI announcement — agent pack, Moody's-tier data partnership, M365 deeper integration — within 90 days is the consensus expectation among industry analysts.
- Microsoft's Copilot positioning. Microsoft's response to a partner's M365 integration outshipping its own Copilot finance experience will be the most-watched dynamic of Q3.
The Planet Tools take
Wall Street is now the second-largest enterprise vertical for frontier-AI deployment after software engineering. Anthropic's May 5 launch is not a feature drop — it is the moment "Claude in finance" became a category, with named customers, ten productized agents, an in-app Moody's, and an M365 surface that out-Microsofts Microsoft. The structural question for Q3 is whether OpenAI and Google can match the named-customer list and the Vals AI score within ninety days, or whether Anthropic locks in Wall Street the same way it locked in software engineering with Claude Code.
Frequently asked questions
What are the 10 Anthropic finance agents released on May 5, 2026?
Pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month-end closer, statement auditor, and KYC screener. Each is shipped as a reference architecture combining skills (workflow markdown), connectors (governed external data access), and subagents (specialized Claude calls).
Which financial institutions are confirmed Claude users?
Anthropic's announcement names Citadel, BNY, Carlyle, JPMorgan, Goldman Sachs, Citi, AIG, Visa, Mizuho, Travelers, Walleye Capital, and Hg as production deployers. Jamie Dimon at JPMorgan went on the record specifically about Claude Code, citing a 20-minute dashboard build with full research backup.
What is Vals AI's Finance Agent benchmark and what did Claude score?
Vals AI's Finance Agent benchmark is an end-to-end test of multi-step financial workflows — pitchbook construction, valuation review, earnings model updates. Claude Opus 4.7 scored 64.37% on the benchmark, which Anthropic positions as industry-leading. Human analysts with two years of experience score in the 60-70% band on the same test.
How does the Microsoft 365 integration actually work?
Anthropic shipped Claude add-ins for Excel, PowerPoint, and Word at general availability, with Outlook in beta. The differentiator versus Microsoft's own Copilot is that a single Claude session keeps full context as the user moves between applications — build a model in Excel, draft slides in PowerPoint, write a memo in Word, all within one continuous reasoning context with no re-prompting required.
What is the Moody's native Claude app?
Moody's embedded its full analytics platform — credit ratings, financials, and risk data on 600M+ public and private companies — into Claude as a native MCP application. Analysts can query the Moody's dataset directly from inside Claude without context-switching to a separate tool. Available at general availability on May 5, 2026.
Are the agents fully autonomous or human-in-the-loop?
Both deployment topologies are supported. In Claude Cowork or Claude Code, the agents run as plugins assisting a human analyst who reviews each step — the default and Anthropic's recommended posture. In Claude Managed Agents (public beta), the agents run more autonomously on Anthropic-hosted infrastructure for workflows like scheduled month-end closes. Anthropic explicitly warns customers to keep humans in the loop "reviewing, iterating on, and approving Claude's work" before any unsupervised deployment.
What other data partners ship with the agent pack besides Moody's?
Anthropic announced MCP connectors for LSEG (Refinitiv) for market data and financial filings, S&P Capital IQ for public-company data and transcripts, Morningstar for mutual fund and ETF data, PitchBook for private-market and venture data, and FIS for anti-money-laundering investigation workflows. Additional vendors are expected to ship Claude-native MCP apps over Q3 and Q4.
Where can a finance team get the agents?
The agents ship via the Claude financial-services marketplace. There are three deployment paths: as plugins inside Claude Cowork or Claude Code subscriptions, as fully managed agents in the Claude Managed Agents public beta, or as reference architectures customers wire against their own Claude API endpoint and connector stack. Pricing is per-seat for the subscription paths and execution-metered for managed agents.
How does Anthropic's launch compare to OpenAI and Google in finance?
OpenAI's enterprise-finance traction has centered on Goldman Sachs's GPT-5.5 rollout earlier in 2026, but without named agent templates or a Moody's-tier data partnership at launch. Google's Gemini Enterprise finance agents announced at Cloud Next 2026 remain in preview as of May. Anthropic's combination of 10 productized agents, full M365 integration, the Moody's app, and a 12-customer named list puts it at least one quarter ahead of either competitor on enterprise finance maturity.
What is Anthropic's revenue trajectory behind this launch?
CEO Dario Amodei disclosed during the launch that Anthropic hit approximately 80x annualized revenue growth in a single recent quarter, well above internal projections. Finance-vertical demand is one of the major drivers behind that curve, alongside Claude Code adoption in software engineering. Anthropic has not broken out a finance-specific ARR number, but Q2 2026 disclosures are expected to be the first to do so.
