Goldman Sachs has pulled advanced AI coding tools from most of its staff. Internal guidance updated in October limits Anthropic's Claude Code and Microsoft's GitHub to the bank's engineers after a review of productivity and risk, Bloomberg News reported. Traders and bankers had used the tools for months to monitor client trading flows, model market positions and assemble written and graphic material. Strats, who mix trading, quantitative research and engineering, keep access. A Goldman spokesperson said the bank makes certain AI tools available to find the best use cases, then hands them to the people best placed to use them long term.
The reversal is sharp. As recently as July, chief information officer Marco Argenti was promoting vibe coding by non-technical employees, with tools built by traders and bankers passed to Goldman's roughly 12,000 engineers to finish. In April the bank removed Claude from its internal platform for Hong Kong bankers, citing its contract with Anthropic, while ChatGPT and Gemini stayed. Goldman's own assistant, which blends models from OpenAI, Anthropic and Alphabet, remains available firmwide. Only third-party agents that write and run code by themselves are fenced off.
There is an awkward overlap. Goldman is one of two lead underwriters on Anthropic's IPO, filed on October 26 for a Nasdaq listing under the ticker ANT, so its bankers will market shares in a company whose coding agent most of them can no longer use. For founders rolling AI agents into production the lesson is blunt: experimentation is easy, but banks carry proprietary data, regulatory obligations and accountability for output. More firms are likely to treat coding agents as a role-based entitlement, not a perk for everyone.