Goldman Sachs has cut off most of its staff from advanced AI coding tools, and not everyone at the bank is pleased. Access to tools like Anthropic’s Claude Code and Microsoft’s GitHub is now limited to engineers, according to internal guidance updated in October and seen by Bloomberg. Traders, bankers and other non-technical employees, many of whom had been using these tools for months, have lost access.The bank says the call came after it looked at productivity and risk. A Goldman spokesperson, Nick Carcaterra, told Bloomberg that the firm offers certain AI tools to staff so it can figure out where they work best. Once that trial period ends, the tools go “in the hands of those who are best suited to leverage them,” he said.There is one exception. The rule does not apply to strats, Goldman’s term for employees who sit between trading, quantitative research and software engineering.Traders seem to feel the change most. According to a person familiar with the matter cited by Bloomberg, some had been using the tools to track client trading flows, model market moves and positions, and put together written and graphic content. On a trading desk, losing that kind of speed is felt quickly.
Goldman’s CIO was cheering vibe coding by traders in July
The reversal stands out because Goldman had been openly encouraging this. In July, chief information officer Marco Argenti spoke about vibe coding by traders and bankers. He described non-technical staff building their own fixes with AI and passing them to one of the firm’s 12,000 engineers to refine.This is also not Goldman’s first AI restriction. Earlier this year, the bank limited the use of Anthropic’s Claude for its Hong Kong staff, though that rule applied only to that location.None of this means Goldman is stepping back from AI. It has built an in-house AI assistant that combines models from OpenAI, Anthropic and Google parent Alphabet. In May, John Waldron, seen as the likely successor to CEO David Solomon, compared the bank to an assembly line run by people and said much of that work could be automated.
Goldman’s AI policy reflects a wider Wall Street rethink
Goldman isn’t the only bank rethinking how its staff use AI. Firms across Wall Street are spending millions of dollars on these tools while also dealing with cybersecurity worries, rising bills and uneven productivity gains. Working out who should use what has become a task of its own.JPMorgan Chase said in July it wanted employees to match models to tasks, so the priciest option isn’t used for something as basic as summarising an analyst report. The idea is to keep costs in check without slowing people down.Goldman’s move follows a similar line, applied to coding assistants. Instead of opening every tool to every employee, the bank is deciding which roles need them and limiting access to those roles.For now, that means engineers and strats. The traders who had built daily workflows around Claude Code and GitHub will have to find another way to get that work done.
