Goldman Sachs has restricted access to Anthropic’s Claude Code and Microsoft’s GitHub to its engineering staff, removing the tools from traders and bankers who had been using them for work. According to Bloomberg News, the bank updated internal guidance in October following a review of productivity and risk concerns.
The restriction marks a significant shift from earlier this year. In July, Goldman’s chief information officer Marco Argenti praised what he called “vibe coding” by non-technical employees, noting that traders and bankers had built tools with AI assistance and passed them to the firm’s engineering teams—approximately 12,000 of Goldman’s roughly 47,000 staff—for refinement.
Traders had leveraged the coding agents to monitor client trading flows, model market moves and positions, and create written and graphic materials. “The change has frustrated some traders,” according to a person familiar with the matter cited by Bloomberg. Goldman spokesperson Nick Carcaterra said the bank makes certain AI tools available “to determine the best use cases” across the firm.
Strategists who work across trading, quantitative research, and software engineering retain access, according to reporting on the matter.
The restriction addresses several risk factors inherent to unreviewed code on trading desks. Claude Code reads files, writes programs, and runs them with the same access as the user. Code written by engineers typically undergoes review, testing, and deployment controls before reaching production systems, whereas trader-built scripts may enter daily use without such oversight. Banks must maintain information barriers between groups—such as between deal bankers and traders who could profit from advance knowledge—and tools traders build independently lack these built-in controls.
Goldman had previously restricted Claude in Hong Kong in April, barring it from the bank’s internal AI platform for bankers there after reviewing its contract with Anthropic. The Financial Times reported the bank read the agreement strictly, and an Anthropic spokesperson noted Claude had never been officially supported in that region.
Despite the restrictions on employee access, Goldman maintains a close relationship with Anthropic. Earlier this year, Anthropic engineers spent six months at Goldman helping build Claude-based agents for trade accounting and client onboarding, which operate in the back office under engineering team management. Goldman and Morgan Stanley are lead underwriters on Anthropic’s planned IPO, expected to list on Nasdaq under ticker ANT before the end of November, according to an S-1 filing from October 26.
Other banks have taken different approaches. Barclays announced in October that it expects half its developers to use Claude Code by year-end, though it simultaneously announced 5,000 job cuts. JPMorgan and Schwab are blocking outside agents from customer accounts, while also tightening access policies.
Key facts
- Goldman Sachs restricted Claude Code and GitHub to engineers in October, following a productivity and risk review
- Traders and bankers had used the tools to monitor client flows, model market positions, and create materials
- The bank’s CIO had praised non-engineer use of AI coding tools as recently as July
- Goldman previously barred Claude in Hong Kong in April based on contract interpretation
- Unreviewed code on trading desks poses risks including information barrier violations and potential execution errors
- Goldman leads Anthropic’s upcoming IPO despite restricting employee access to the company’s coding tool