Goldman Sachs has pulled advanced AI coding tools from most of its traders and bankers, keeping them for its engineers only, according to Bloomberg's reporting on internal guidance. The affected tools reportedly include Anthropic's Claude Code and GitHub's coding assistant. The move reverses a policy the bank had promoted just a few months earlier.
What changed
Bloomberg reported that guidance issued in October limits the tools to the bank's engineers after a review of productivity and risk concerns. Traders and bankers had been allowed to use them for months. According to reports, some had used them to monitor client trading flows, model market changes and positions, and create written and graphic materials, and some are frustrated by the change.
A Goldman spokesperson told Bloomberg that the bank makes certain AI tools available to find the best uses across different parts of the firm. The company has said the experimentation phase is over and that it is limiting the tools to the people best suited to use them over the long term, as reported by The Decoder. Goldman cited "productivity and risk management" among the factors but did not give details.
A reversal in months
In July, Goldman's chief information officer, Marco Argenti, spoke favourably of "vibe coding" by non-technical staff. The idea was that a trader or asset manager could build a quick prototype with AI and hand it to one of the bank's roughly 12,000 engineers to turn into something robust. In August, a Goldman partner leading a flagship AI project, Chris Churchman, warned that heavy reliance on AI across Wall Street could weaken the reasoning skills of the next generation of bankers. Earlier this year, Goldman also restricted its Hong Kong bankers from using Anthropic's Claude.
Some coverage reports that affected staff are being directed to the bank's own GS AI Assistant or told to request alternative tools. That detail comes from secondary reporting and was not in the Bloomberg excerpts reviewed.
What it signals
The point is less that Goldman is abandoning AI and more that it wants control over who uses which tools and for what. Banks handle sensitive data, face heavy regulation and must be able to explain how code and models that influence decisions were produced. An employee with no engineering training can now generate working software in minutes, but cannot necessarily review, test or secure it. For risk and compliance teams, that is a new category of exposure.
Other large organisations, including banks and telecoms firms across Africa that are adopting AI tools, may face the same question: should powerful coding agents be open to everyone, or only to people trained to check their output? Many are likely to settle on role-based access, with approved tools, logging and review rules. Whether other banks follow Goldman's lead is the thing to watch.
Anthropic, the maker of Claude Code, is the company behind this assistant. This article reports Bloomberg's account and Goldman's statements, and Anthropic had not been quoted in the coverage reviewed.
Sources: Bloomberg, via Quartz, The Decoder, Finextra and PYMNTS; Cryptobriefing; Wisevoter; DEV Community.
