On November 9 Anthropic wrote that JPMorganChase is rolling Claude out across the firm, to employees beyond engineers. In the Commercial and Investment Bank, work reaches production faster, and the bank redeploys the freed-up capacity against its own priorities.
Who is using it for what
Engineers are modernizing legacy systems and building their own agents, including pull request review agents and self-healing agents, and sharing what works. The security team uses Anthropic’s frontier models to find and fix vulnerabilities, work Anthropic has described before.
The breadth is the new part. Product teams turn requirements straight into engineering backlogs, and other teams use Claude Code for risk and controls assessments and program planning. In Asset & Wealth Management the bank puts the people with the deepest business insight into small teams alongside technology and product experts, to build customized client investment solutions faster. Last month J.P. Morgan Asset Management also announced a connector that brings 55ip’s ActiveTax technology for tax-aware portfolio transitions into advisors’ daily work.
The guardrails are the interesting part
Claude Code runs at JPMorganChase in controlled environments, including Devspace, the firm’s containerized development platform, on its own infrastructure. The intent is to keep agents from getting direct access to employee credentials or internal systems. Access widens in stages, and every new Claude model goes through the firm’s own approval process before anyone there can use it.
Code ships under the existing discipline: thousands of automated tests and approval gates check every change, and an automated gate returns a hard pass or halt that the pipeline enforces. Claude cannot sign off on its own work. People stay accountable for every production change.
Two banks, two directions
The announcement lands a week after a different move on Wall Street. According to a Bloomberg report from November 2, based on internal guidelines from October, Goldman Sachs cut off most employees from advanced AI coding tools including Claude Code and limited them to engineers. A spokesperson gave “a variety of factors including productivity and risk management” as the reason and said the experimentation phase was over.
JPMorganChase is going the other way, and it uses the guardrails as the argument. Read side by side, the two cases are two verdicts on one question: whether agents can be fenced in inside a regulated environment before you hand them to everybody. Goldman answered no. JPMorganChase is building its answer right now.
This is still a post by Anthropic about its own customer. There are no numbers on how many people use it, on error rates, or on changes the gates rejected, and until those exist it describes a setup rather than a result.