What happened
Bloomberg reported on November 2, 2026 that Goldman Sachs has restricted most traders and bankers from using AI coding tools, including Anthropic's Claude Code and Microsoft's GitHub. According to internal guidance updated in October and seen by Bloomberg, the tools are now limited to engineers. Bloomberg and the outlets that followed it say the bank had allowed bankers to use the tools for months, and that the change has frustrated some traders. They had used the tools for things like following how clients were trading, simulating shifts in markets and positions, and drafting text and visuals. Quartz, relaying Bloomberg, adds that the directive does not cover "strats," a group whose jobs combine software engineering, quantitative research, and trading. Goldman spokesperson Nick Carcaterra told Bloomberg: "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." The reporting comes from Bloomberg and its guidance source. Goldman has not published the policy itself, and the specific risks behind the decision have not been spelled out.
Why it matters for business owners
In July, Bloomberg reports, Goldman's chief information officer spoke favorably about non-technical staff building their own tools with AI and handing the results to an engineer to polish. Three months later the practice is narrower. That is the part worth noticing. A company with far more resources than yours ran a period of open experimentation, looked at productivity and risk, and drew a line. Access was a decision, and it was revisited. Most smaller businesses never make that decision. Someone signs up for a coding assistant, a teammate copies the setup, and a few months later several people are building scripts, spreadsheets, and automations nobody reviewed. Nobody chose that. It accumulated.
What owners should not misunderstand
Do not read this as proof that AI coding tools are unsafe. Goldman has not said they are, and the reporting does not describe an incident. The stated reasons are productivity and risk management, nothing more specific. Do not read it as proof the tools failed either. Engineers keep access, and the bank has its own AI assistant that draws on models from OpenAI, Anthropic, and Alphabet, according to Quartz's summary of Bloomberg. The bank narrowed who uses one category of tool. It did not step away from AI. Do not assume a bank's rule fits your company. Goldman has 12,000 engineers, according to Bloomberg's reporting as relayed by PYMNTS. It can send a trader's prototype to a team that reviews and hardens it. A thirty-person firm may have one developer, or none. The same rule could mean something very different for you. And do not assume the opposite mistake, that restricting access solves the problem. A ban with no alternative tends to push people toward personal accounts and unapproved tools. Our earlier post on shadow AI covers that pattern.
The operational lesson
A tool that writes code changes who can build software inside your company. That is the real shift. Someone in sales or finance can now produce a script that touches customer data, a spreadsheet macro that changes numbers, or an automation that sends messages. That can be useful. It also means work product with no reviewer, no owner, and no record of what it does. The question is not whether the output looks right on the day. It is who maintains it, who checks it, and what it can reach. So the access decision has two parts. First, who should be able to create things with the tool. Second, what has to happen before something they create is relied on. A prototype used by its author is a different thing from a script that runs on customer records every night.
What a serious business should do next
List the AI tools people use for work, including coding assistants, and who uses each. Ask, do not guess. You will likely find more than you bought. Sort the use into three groups: personal drafting and research, building or changing tools and automations, and anything that touches customer data, money, or records. Each group needs a different level of care. For anything people build with AI, name an owner and a reviewer before it is relied on. Even a short note on what it does, what it can access, and who answers for it helps. Decide what data may go into which tool. A coding assistant pointed at a folder of customer files is a data decision, not just a tooling decision. Offer an approved route. If you narrow access, say where people can go instead. A clear option beats a quiet workaround. Revisit the decision. Goldman's own sequence shows that early access is not permanent. Put a review date on the calendar, and judge by real use, not by how a tool looked in a demo.
The Atlacis view
Atlacis helps owners slow down, understand the workflow, map the risk, and choose the right tool or system before spending. Who gets which tool is part of that map, and it is much easier to settle before habits form than after. If you are not sure who in your company is using AI to build things, or what those things can reach, a short conversation can help you see it clearly before you set a policy.
The short version
- Bloomberg reports Goldman Sachs limited Claude Code and GitHub coding tools to engineers, after months of wider use by bankers and traders.
- Goldman cites productivity and risk management. No specific incident has been reported.
- Engineers keep access, and the bank has not stepped away from AI. It narrowed who uses one kind of tool.
- AI coding tools let non-engineers create software, so the question becomes who reviews and owns what they build.
- Sort uses by risk, name an owner and reviewer for anything built with AI, and offer an approved alternative when you restrict a tool.
- Treat access as a decision with a review date, not a default.