Goldman Sachs restricts bankers' and traders' use of Claude Code and GitHub
Goldman Sachs pulled AI coding tools from most bankers and traders, keeping them for engineers and strats.
Objective Facts
Goldman Sachs has restricted use of AI coding tools, including Anthropic's Claude Code and Microsoft's GitHub, to its engineers under internal guidance updated in October, Bloomberg reported on November 2, 2026. Bankers and traders had been allowed to use the tools for months, and some traders were frustrated because they relied on them to track client trading flows, model market and position changes, and produce written and graphic content, according to the Bloomberg story as republished by The Edge Malaysia. Goldman spokesperson Nick Carcaterra said the firm makes AI tools available to find the best use cases, then assigns them to those best suited to use them long term. Staff known as strats, who work across trading, quantitative research and engineering, are not covered by the restriction. Goldman has not disclosed what its internal review found, and no outlet has reported a specific incident behind the decision.
Deep Dive
Goldman spent much of 2026 promoting AI inside the bank. In July, Marco Argenti praised traders and bankers who used tools such as Claude Code to build rough prototypes in plain language and hand them to the firm's roughly 12,000 engineers, according to the Bloomberg report carried by The Edge Malaysia. The bank also built an internal assistant that draws on models from OpenAI, Anthropic and Alphabet, and in May chief operating officer John Waldron called Goldman a "human assembly line" with room for automation. The October guidance narrows the most powerful coding agents back to engineers and strats. Goldman had already cut Claude access for Hong Kong bankers in April, a move Reuters said came as banks increased scrutiny of AI tools over data security and cyber risk. The stated reason is thin. Carcaterra cited productivity and risk management, and FStech and Techstrong.ai both reported the decision followed an internal review, but Goldman has not said what it found. Techstrong.ai listed possible concerns such as protecting material non-public information, keeping information barriers intact and auditing machine-generated code, while noting these were broader industry worries. DiarioBitcoin pointed out that no reported incident, coding error or loss is tied to the decision, and Gadget Review's Al Landes wrote that the restriction is well sourced while the cause is not. James Harrison, chief executive of Genesis Global, told Techstrong.ai that AI-built software is safe only with engineers unless business teams get foundations IT already trusts. Coderhouse's Francisco Rhaiel argued the episode shows generating code is easy and answering for it is hard, though his company sells programming courses. There is also a minor discrepancy over the tools: Bloomberg's account names GitHub, while AI Weekly specified GitHub Copilot. Open questions include how many people lost access, whether the change is permanent and whether non-engineers will get a sanctioned way to propose prototypes, gaps DiarioBitcoin flagged. The CSR Journal cited Replit's chief executive predicting on X that a wider crackdown in finance is coming, a claim no outlet has tested. Goldman's ties to Anthropic add another layer: Bloomberg and Techstrong.ai reported the bank is lead underwriter on Anthropic's planned IPO, so any further shift in how Goldman uses Claude will draw attention. JPMorgan's July guidance urging staff to match AI models to task complexity suggests other banks are also tightening control of AI spending and use.