The bank is limiting advanced AI coding tools like Anthropic's Claude and Microsoft GitHub mainly to engineers, and some traders are not happy about it
Goldman Sachs is pulling some of its shiniest AI toys off the trading floor.
Under updated internal guidelines announced in October 2026, the bank is restricting access to advanced AI coding and programming tools, mainly to its engineering staff. Traders and bankers who had grown used to those tools now find the door partly closed.
The affected tools include Anthropic’s Claude and Microsoft GitHub. Both had been available to non-technical employees. Going forward, they will primarily serve engineers.
What changed, and who got left out
The core of the policy is simple. If you write code for a living at Goldman, you keep your AI coding assistants. If you mostly trade or advise clients, your access is being narrowed.
Traders had been using these tools for work like trade monitoring, building models and generating content. The change has frustrated traders who leaned on these tools for that work.
There is one notable carve-out. Quant strategists are exempt from the new restrictions.
Their jobs straddle trading and engineering, mixing market work with heavy technical building. Goldman appears to have concluded that cutting them off would make little sense when coding is already part of the role.
Goldman’s explanation is measured. A spokesperson for the bank said the adjustments came after a trial period that weighed the efficiency gains from these tools against risks and controls.
A pattern of tightening, not a one-off
This is not the first time Goldman has narrowed AI access in 2026.
In April, the bank restricted access to Anthropic’s Claude for bankers based in Hong Kong. That move followed a contract review.
Taken together, the two decisions sketch a firm that is becoming more selective about which outside AI tools go where, and to whom. The April move was geographic. The October move is functional, sorting staff by job type rather than location.
Goldman is still rolling out its proprietary GS AI Assistant to approximately 46,000 employees. It is also running ongoing trials of agentic AI, including in trade surveillance and other areas.
So the message is less “no AI” and more “our AI, on our terms.” Third-party coding tools get fenced in. The in-house assistant gets handed out widely.
