Anthropic Targets Wall Street: 10 FSI Agents and $1.5B Goldman JV
Anthropic launched 10 production-ready Claude agent templates for financial services on May 6, backed by a $1.5B joint venture with Goldman Sachs and Blackstone — a pivot from API vendor to operating layer in one of the world's most regulated sectors. Three independent sources reported the same coordinated launch: AlphaSignal's lead story, Anthropic's own GitHub repo hitting #1 on GitHub Trending, and an X news ticker, with Jamie Dimon appearing on stage at the event.
What the Source Actually Says
The ten named agents — Pitch Agent, Market Researcher, GL Reconciler, KYC Screener, Earnings Reviewer, Model Builder, Month-End Closer, Valuation Reviewer, Statement Auditor, and Meeting Prep Agent — each ships with domain-specific skills, slash commands (/comps, /dcf, /ic-memo), and live access to 11 MCP-connected paid data providers: Daloopa, Morningstar, S&P Global, FactSet, Moody's, MT Newswires, Aiera, LSEG, PitchBook, Chronograph, and Egnyte. No custom engineering is needed to install and run them.
The critical architectural decision in the public GitHub repo is dual deployment: the same system prompt and skill set runs either as an interactive Claude Cowork plugin or as a headless Managed Agent via /v1/agents for overnight autonomous runs — one codebase, two surfaces. The repo also includes Microsoft 365 add-in tooling routing Claude inside Excel, PowerPoint, Word, and Outlook against a firm's own cloud (Vertex AI, Bedrock, or an internal LLM gateway), removing data-residency friction for enterprise procurement.
Every output carries a strict HITL gate: the repo's disclaimer states "agents do not bind risk, post to a ledger, or approve onboarding — every output is staged for human sign-off." On the same day, Anthropic published a safety paper documenting a training technique that cut unsafe agentic behavior from 54% to 7% — timed, as AlphaSignal noted, to coincide with overnight autonomous agent deployments at Goldman Sachs.
Strategic Take
The dual-deployment model and 11 MCP-standardized data connectors make the stack portable across clouds and firm-customizable without re-engineering. Builders in regulated verticals should treat Anthropic's mandatory HITL output gate as the procurement checkpoint that gets compliance to sign off — not a legal footnote but the pattern every enterprise AI deployment in finance will need to adopt.
