Goldman Sachs limits Claude Code and GitHub Copilot to engineers, cutting off traders and bankers

Goldman Sachs has pulled back access to advanced AI coding assistants for most of its staff. Bloomberg reported on November 2 that internal guidelines issued in October now reserve tools such as Anthropic's Claude Code and Microsoft's GitHub Copilot for the bank's engineers, leaving traders and bankers without them. The Decoder and PYMNTS have relayed the report; Goldman has not published the guidance itself.

From encouragement to restriction

The move marks a turn from the bank's earlier message. In July, chief information officer Marco Argenti spoke approvingly of "vibe coding" by non-technical employees. The idea he described was that someone on a trading desk or in asset management would put together a quick AI-built prototype, then hand it to one of the firm's roughly 12,000 engineers to turn into production software. According to The Decoder, that workflow is no longer the plan.

For several months, people outside engineering had used the tools in their daily work. A source told Bloomberg that some traders had relied on them to monitor client trading flows and to produce written and visual materials, and that the cutoff has caused frustration.

Goldman's explanation

Spokesperson Nick Carcaterra said the bank weighed a "variety of factors including productivity and risk management" during an initial period of testing. He described that experimentation phase as finished and said the tools will now go to the staff best placed to use them over the long term. Goldman has not given figures on how many employees lost access, what the tools cost, or whether specific incidents prompted the decision.

Banks set tighter rules for AI

The decision comes as large lenders spend heavily on AI while trying to contain cost and security risks. PYMNTS notes that JPMorgan chief executive Jamie Dimon has said outside AI agents will not be allowed inside the bank, that Bank of America limits agents' access to consumer deposits, and that Citigroup does not let third-party agents log into accounts or carry out transactions.

Goldman's case is somewhat different: it concerns tools the bank itself had approved, not outside agents. The open question is whether coding assistants in the hands of non-engineers create code and data risks that outweigh the time they save, or whether this is mainly a cost and governance reset. Goldman's own statement points to both productivity and risk without saying which weighed more.