Goldman Sachs has limited Claude Code and GitHub to engineers after allowing bankers to use the tools for months, Quartz reported on 2 November. Internal guidance updated in October changed access for traders and bankers at the investment bank. The restrictions have annoyed some employees, Bloomberg reported on 3 November.
Key points
- October guidance limits Anthropic’s Claude Code and Microsoft’s GitHub to engineers after a review of productivity and risk concerns.
- Some traders had used the tools to follow client trading, model market changes and prepare text and visuals.
- The restriction does not cover strats, whose work combines engineering, quantitative research and trading.
- In July, Goldman’s chief information officer had praised AI-assisted coding by staff outside technical roles.
October guidance limits Claude Code and GitHub
The guidance confines Anthropic’s Claude Code and Microsoft’s GitHub to engineers following a review of productivity and risk concerns. Bankers had been permitted to use the tools for months before the change. Some traders were frustrated by the restriction, having used the tools to track client trading activity, simulate changes in markets and their positions, and prepare written and graphic material.
Goldman’s directive exempts strats — employees whose roles span software engineering, quantitative research and trading. Their continued access leaves a group involved in trading work able to use tools that other traders had employed for market analysis and client-flow monitoring.
Goldman spokesperson Nick Carcaterra said the bank makes tools available to employees to identify uses across its businesses. “After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term,” he said, according to Quartz.
That approach follows a period in which bankers could build with the tools themselves. The October guidance changes who can use Claude Code and GitHub, rather than ending the bank’s use of AI: Goldman has also created an internal assistant drawing on models from OpenAI, Anthropic and Alphabet.
Argenti’s July approach drew on 12,000 engineers
In July, Goldman Chief Information Officer Marco Argenti spoke favourably of “vibe coding” by employees outside technical roles. Traders and bankers used AI to develop solutions before passing them to an engineer from the bank’s pool of 12,000 for refinement, he said. The October guidance removes access to the coding tools from many of the employees who had supplied those initial solutions.
Goldman has made a separate investment in AI beyond staff coding tools. AI Affairs previously reported that the bank launched an AI institute within a $500 million expansion. Earlier in 2026, Goldman restricted staff in Hong Kong from using Anthropic’s Claude, Quartz reported.
JPMorgan and Citigroup set agent limits
Other banks have drawn boundaries around different AI uses. JPMorgan Chase chief executive Jamie Dimon warned of “rogue agents” during an October earnings call and said outside AI agents would not be permitted inside the bank, PYMNTS reported. Goldman’s October decision concerns staff access to coding tools, rather than outside agents entering the bank.
Bank of America has placed limits on agents’ access to consumer deposits while developing rules for liability when an AI authorises a payment, according to PYMNTS. Citigroup does not allow third-party agents to sign into accounts or make transactions. Those restrictions concern access to customer accounts, a different operational decision from Goldman’s allocation of coding tools among employees.
