BUSINESSCompanies · · 2 min read · Source: Bloomberg
Goldman Sachs limits AI coding tools to engineers
Goldman Sachs has removed most employees' access to advanced AI coding tools such as Claude Code, Bloomberg reported on November 2, 2026. Only engineers can still use them.
Key facts
- Bloomberg reported on November 2, 2026 that Goldman Sachs is limiting AI coding tools like Claude Code to engineers.
- Traders and bankers are mainly affected; the internal guidelines date from October.
- The bank cites productivity and risk management and says the experimentation phase is over.
What changes at Goldman
According to Bloomberg, the restrictions stem from internal guidelines issued in October. Traders and bankers are the main group affected. They had been using tools like Anthropic's Claude Code and GitHub's AI tools for months, for example to monitor trading flows or create content. Access is now reserved for engineering teams, which reportedly frustrated some staff.
The reasoning
A bank spokesperson cited several factors, including productivity and risk management. The experimentation phase is over, he said, and access now goes to the people best placed to use the tools over the long term. The bank did not go into more detail.
The shift is notable. As recently as July, the bank's CIO had praised non-technical staff building small programs with AI. The plan at the time: business teams build a prototype, engineers make it production-ready. According to the report, that approach no longer applies.
Not the first move of its kind
In April, Goldman had already cut off Claude access for staff in Hong Kong. Reuters and Bloomberg reported that this was about data security and contractual issues in that region. The new step is different: it is not about one location but about who in the company should build software with AI at all.
That is a signal for other companies. Many rolled out AI tools broadly first to see what would happen. Some are now moving into a phase of sorting out rules, ownership and costs.
What this means for you
- Large companies are moving from "everyone can experiment" to clear rules on who uses which AI tool.
- Small home-built programs from business teams are becoming a risk question: who maintains them, who is accountable?
- For employees, what is allowed today can disappear tomorrow through an internal policy.
How to use this
Write a short rule for your team: which AI outputs can business staff use on their own, and when does someone with technical responsibility need to review them? Half a page is enough to start.
Luigi's take
I don't read this as a rejection of AI. It is what comes after every big trial phase: cleaning up. When traders build their own small programs, you quickly end up with tools nobody reviews and nobody maintains. A bank can afford that less than most. It is still a pity, because a lot of knowledge from the business side gets lost along the way.
My advice for companies: don't ban across the board. Define what business teams may build themselves (analyses, drafts, prototypes) and when an engineer has to take over (anything that touches customer data or runs permanently). Put that line in writing before the first problems show up.
Source
Bloomberg: Goldman restricts traders and bankers from using some AI tools
More sources
- The Decoder – Zusammenfassung des Bloomberg-Berichts mit Stellungnahme von Goldman Sachs
This article was researched and written with AI assistance. Responsible: Luigi Lugmayr.
