Goldman Sachs has barred most of its traders and bankers from using AI coding tools, including Anthropic's Claude Code and Microsoft's GitHub, reversing months of open experimentation at the bank, Bloomberg reports.

Internal guidance updated in October restricts the tools to engineers. The bank had allowed bankers to use them for months before a review of productivity and risk concerns forced the change.

Who loses access, and who keeps it?

The restriction has frustrated some traders, who had built the tools into their daily workflow. They used them to track how clients were trading, simulate shifts in markets and in their own positions, and draft text and visuals — work that now returns to manual methods or to the engineering queue.

The directive does not cover "strats," the hybrid role that combines software engineering, quantitative research and trading. For practitioners, that carve-out matters: if your job already includes writing production code, you keep the tools.

Goldman framed the move as the natural end of a testing phase. "We make certain AI tools available to our people to determine the best use cases across different parts of the firm," Goldman spokesperson Nick Carcaterra said. "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."

What changed since summer?

The October guidance marks a sharp turn from July, when Chief Information Officer Marco Argenti spoke favorably of "vibe coding" among employees outside technical roles, traders and bankers included. Argenti described a pipeline in which those employees used AI to build their own solutions, then handed the results to one of the bank's 12,000 engineers for polishing.

A month later, a senior voice inside the bank flagged the longer-term risk. Partner Chris Churchman, who co-chairs Goldman's Global Banking and Markets AI working group, warned in August that the spread of AI across Wall Street risks degrading the analytical abilities of the next generation of bankers. He said Goldman had not yet determined how it would manage that transition.

October's answer, in effect: confine the code-writing tools to the people already trained to judge what the code does.

How does Goldman's approach compare with rivals?

Other banks are drawing their own lines. JPMorgan Chase said in July that it wants staff to match each task to a suitable model, reserving the priciest option for harder work rather than something like condensing an analyst report. The logic parallels Goldman's — control cost and risk by routing usage, not banning it outright.

Goldman has also hedged its dependence on outside vendors. The bank built its own AI assistant, which draws on models from OpenAI, Anthropic and Alphabet, giving it an internal layer it can govern directly.

This is not Goldman's first restriction. Earlier this year the bank blocked staff in Hong Kong from using Anthropic's Claude, so the October directive extends an existing pattern rather than starting one.

For anyone running AI policy at a financial firm, the signal is straightforward: pilot broadly, then narrow access to the roles where oversight is native — and expect the traders who lost their shortcuts to notice.