Goldman Sachs Limits AI Coding Tools to Its Engineers

What happened

Goldman Sachs has restricted the use of advanced AI coding tools to its engineers, according to a Bloomberg report published on 2 November 2026. The bank, which is based in New York, had for months let bankers use Claude Code and GitHub for work. Bloomberg says internal guidance updated in October now limits those tools to engineers. The report says the review weighed productivity and risk concerns.

Some traders are unhappy with the change. According to Bloomberg, they had used the tools to follow client trading flows, model shifts in markets and positions, and produce written and graphic material. A person familiar with the matter told Bloomberg that the restriction has frustrated some traders. Goldman's spokesperson, Nick Carcaterra, said the bank makes certain AI tools available so it can find the best use cases across the firm. Per The Decoder, which cites Bloomberg, Carcaterra also said the experimentation phase is over and that access now goes to the people best placed to use the tools over the long term.

The bank's in-house tool is the part of the story that matters most for general staff. Wisevoter reports that Goldman is rolling out its GS AI Assistant to about 46,000 employees, and that it is running trials of agentic AI, including in trade surveillance. Finextra notes that Goldman had already restricted access to Claude for staff in Hong Kong earlier in 2026, amid rising tension between the United States and China over AI.

Who keeps access and who loses it

The reporting gives a clear split. The table below sets out the groups named in the coverage.

Neither Bloomberg nor Goldman has said how many staff used the coding tools, what the bank pays for them, or how the vendors charge. Those figures are not public, so the size of the change inside the bank cannot be measured from the reporting.

Why it matters for businesses that automate work

In July, Goldman's Chief Information Officer, Marco Argenti, praised what he called vibe coding by non-technical staff. Bankers and traders built prototypes with AI tools and passed them to engineers to improve. That route is now closed for most non-engineers. The reversal shows that a fast way to build tools can create a control problem once the tools touch trading or client data.

Finextra notes that AI-generated code may need careful testing and review before it is used in systems that handle financial data or trading activity. That point applies to any business, not only banks. A spreadsheet macro or a small script built by an operations clerk can move money, change a ledger or send a wrong report to a customer. The risk is not the tool alone. It is the absence of a named person who checks the output.

What changes in practice for a business that automates work

Goldman's move is one bank's internal policy, and the reporting does not name a regulator rule behind it. Still, the steps it points to are useful for a firm of any size, from five people to several thousand.

1. List every AI tool staff already use, and write down what each person does with it.
2. Separate general assistants, used for drafting and summarising, from tools that write or change code or spreadsheets.
3. Decide which tools may touch live customer, payroll, billing or ledger data, and limit those to named owners.
4. Require a technical or finance owner to review any prototype before it runs on real data.
5. Keep a record of who approved each automated step, so the process can be audited later.

Restricting access has a cost. Staff who built useful fixes may lose a way to solve their own problems, and a small firm may find that a review process costs more time than the work it saves. For a rare, one-off task, a manual process is often the better choice. Automation pays off when the same routine repeats and the source data is clean enough to check.

What to watch next

Three things will show whether this becomes a wider pattern. The first is whether other large banks adopt role-based access to coding tools, as the reporting suggests they are still deciding how staff should use AI. The second is whether Goldman publishes more detail on the GS AI Assistant and on its agentic trials. The third is whether the work traders lost is moved to approved tools, or simply stops. Each answer will tell a business owner how far the trend goes beyond one bank.