Goldman Sachs Pulls AI Coding Tools From Traders and Bankers, Limiting Them to Engineers

Traders and bankers had been using the tools for months to monitor trading flows and produce content, the reports said. Bloomberg reported that the change has frustrated some employees inside the firm.

The Bank's Explanation

Goldman spokesperson Nick Carcaterra told Bloomberg the decision reflected a "variety of factors including productivity and risk management." He said the period of experimentation was over and the firm was now restricting the tools to the people "who are best suited to leverage them for the long term." The bank did not give further detail on which risks or productivity concerns drove the change.

The move is a reversal of the approach set out earlier this year. In an interview in July, Goldman's chief information officer, Marco Argenti, spoke positively about "vibe coding" by non-technical staff, according to The Decoder. He described a process in which a trader or asset management employee would turn an idea into a rough prototype with AI and then pass it to an engineer to build into a production system. Under the new guidelines, that route for non-engineers appears to be closed.

Not the First Restriction

This is not the first time Goldman has narrowed staff access to AI tools. In April, Reuters and the Financial Times reported that the bank had removed access to Anthropic's Claude models for its Hong Kong-based bankers. Sources described that decision as following a strict reading of Goldman's contract with Anthropic rather than a regulatory requirement. Anthropic told the FT at the time that Claude had never been officially supported in Hong Kong.

Why It Matters

The new restriction is broader. It applies to a type of tool, not a location, and it comes from one of the Wall Street banks that has been most public about using AI. For other financial institutions, it highlights the governance problem of letting non-engineers build software. Code written quickly by traders or bankers can touch sensitive data and trading systems, but it may not go through the testing, review and audit trails that regulated firms expect for production software.

The decision also matters for AI vendors. Coding assistants have been among the fastest-growing enterprise AI products, and companies such as Anthropic and Microsoft have promoted their use well beyond professional developers. A large bank deciding that those tools should stay with engineers, at least for now, suggests some enterprise buyers are moving from broad trials to narrower, controlled deployments.

Bloomberg's report did not say how many Goldman employees lost access, or whether the bank plans to bring the tools back to some non-engineering roles later.