Goldman Sachs Reportedly Takes AI Coding Tools Away From Most Traders and Bankers

Goldman Sachs handed its traders and bankers some of the most powerful AI coding tools on the market. Now it wants them back. According to a Bloomberg News scoop published Monday, the bank has limited Anthropic's Claude Code and Microsoft's GitHub tools to engineers, after months in which bankers were free to use them.

The change appears in internal guidance updated in October and seen by Bloomberg. It is not a firm-wide ban. Engineers keep access, and so do "strats," the hybrid group whose jobs mix software engineering, quantitative research and trading, according to the report as relayed by Quartz. Most traders and bankers are out.

The bank's explanation says very little

Goldman spokesperson Nick Carcaterra gave Bloomberg a statement that explains the process and skips the specifics.

"We make certain AI tools available to our people to determine the best use cases across different parts of the firm. 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." Nick Carcaterra, Goldman Sachs spokesperson

Read that twice. It names productivity and risk management as factors, and nothing more. The reporting does not say what the risks were, whether a specific incident prompted the review, or what productivity numbers came out of it. Goldman has not detailed any of that, and nobody should fill the gap with guesses.

What the reporting does describe is how the tools were used. Bloomberg says traders turned to them to follow how clients were trading, to simulate shifts in markets and in their own positions, and to draft text and visuals. That is real desk work, and some traders are frustrated at losing it, a source familiar with the matter told Bloomberg, per PYMNTS.

One caveat. Only Bloomberg has seen the guidance. Quartz, PYMNTS and The Decoder relay its reporting rather than confirm it, so treat the details as one outlet's account of an internal document.

The reversal is what makes this interesting. In July, Goldman's chief information officer, Marco Argenti, spoke favorably of "vibe coding" by employees outside technical roles, including traders and bankers, according to Bloomberg. His description: a non-engineer builds a solution with AI, then passes it to one of the bank's 12,000 engineers to polish. Three months later, the first half of that pipeline is shut for most of the people who were running it.

My read, and it is only a read, is that Goldman is behaving like a cautious institution. Let people experiment, then pull the tools back to a smaller group. A trader with a code assistant can ship something fast. A bank has to be able to explain, audit and support whatever gets shipped. Whether that is what happened here, the bank has not said.

There are other signs of unease. Earlier this year Goldman restricted staff in Hong Kong from using Claude, Bloomberg reported. In August, partner Chris Churchman, who co-chairs the bank's Global Banking and Markets AI working group, warned that AI's spread across Wall Street risks degrading the analytical abilities of the next generation of bankers. He said Goldman had not yet decided how to handle that. Goldman's economists also forecast AI job displacement, but that is a separate story.

Goldman is hardly turning its back on AI. Bloomberg notes the bank has its own AI assistant, which draws on models from OpenAI, Anthropic and Alphabet. Rivals are still setting their own rules too. JPMorgan said in July that it wanted staff to match each task to a suitable model, saving the priciest option for harder work.

The industry mantra has been an AI tool for every employee. Goldman's answer, at least for coding tools, is that access follows the job. Bankers who spent the summer building their own tools will want to know whose call that was, and so far the bank has offered one paragraph in reply.