Goldman Pulls AI Coding Tools Back

Goldman Sachs has barred traders, bankers and other non-engineering staff from using advanced AI coding tools, according to Bloomberg, a retreat from the broader internal AI use its technology leadership had appeared to encourage.

Fast Company, citing Bloomberg, reported that access to tools including Anthropic’s Claude Code and Microsoft’s GitHub has been restricted to engineers and “strats,” Goldman’s quantitative engineering and analytics specialists. The reported reason was not some grand philosophical revolt against AI, but the more familiar corporate mix of risk control and productivity concerns.

That matters because the tools were reportedly being used for practical office work: monitoring data, building financial models and producing graphics. In plain English, Goldman is drawing a line between employees who write production-grade code and employees who merely found that code-adjacent tools can save time. In banks, “merely” tends to be where compliance departments start reaching for the brakes.

The move also appears to cut against remarks Goldman chief information officer Marco Argenti made in July, when he praised nontechnical staff for using AI to build software prototypes before handing them to engineers for refinement. A policy can change without anyone admitting the earlier enthusiasm was overdone. Large firms do that all the time.

The report leaves some important details unclear, including whether the restriction is temporary, how broadly “advanced” tools are defined, and whether Goldman plans a narrower approved path for nontechnical staff.