FreshSynapse

AI trends & daily updates, in plain English

AI Business

Goldman Sachs Pulls Advanced AI Coding Tools From Bankers and Traders

By FreshSynapse Editorial · · 3 min read

A businessman in a suit checking his phone outside an office building

Photo: Hippopx (public domain).

What happened. Goldman Sachs has restricted advanced AI coding tools to its engineering staff, ending months in which bankers and traders were free to experiment with them. Bloomberg reported the change on 2 November 2026, citing internal guidance updated in October.

Under the new rules, Anthropic’s Claude Code and Microsoft’s GitHub tools are generally reserved for engineers — the bank has about 12,000 of them, roughly a quarter of its workforce. “Strats,” staff who sit between trading, quantitative research and software engineering, are exempt and keep access. Bankers and traders who had been using the tools lose them.

The reversal. The reversal is striking because of how recently the bank celebrated the opposite approach. In July 2026, chief information officer Marco Argenti publicly praised “vibe coding” — non-technical staff building quick prototypes with AI and handing them to an engineer to bring into production. For months, traders and bankers used the coding tools for work far from programming: monitoring client trading flows, modelling market moves and positions, and producing written and graphic content. Some traders have objected to losing access, according to the reporting.

Goldman spokesperson Nick Carcaterra attributed the decision to “a variety of factors including productivity and risk management,” adding that the experimentation phase is over and the tools will stay with those best placed to use them over the long term. No specific incident, cost figure or internal audit behind the decision has been made public, and the full internal guidance has not been published. The bank has not banned AI itself: it continues to run its own GS AI Assistant, launched firmwide in June 2025, which aggregates models from OpenAI, Anthropic and Alphabet.

Why it matters beyond one bank. The story is not, as some headlines suggest, a bank turning against AI. It is a large employer drawing a line that many companies are now wrestling with: who should be allowed to use powerful tools, for what, and with what oversight. Coding agents that can write, refactor and run code are genuinely useful in untrained hands — and genuinely risky there too, since the user may not be able to judge whether the output is correct, secure or compliant.

Goldman has form here. In April 2026, Bloomberg reported, the bank separately restricted Claude access for staff based in Hong Kong — a location-based limit, where this one is role-based. The pattern suggests a company moving from broad experimentation to narrower, governed deployment: AI for everyone in approved general-purpose forms, powerful coding agents only for the people trained to check their work.

For other workplaces watching, the message is practical rather than dramatic. The “let everyone try everything” phase of workplace AI is ending at big firms, and the next phase is job-specific access with named owners. The experiment was real, and so is the pullback — both are part of the same learning curve.

Sources

  • Goldman Sachs
  • Claude Code
  • GitHub Copilot
  • AI coding
  • banking
  • workplace AI

Related articles