Goldman Sachs has cut most of its staff off from advanced artificial intelligence coding tools, restricting tools like Anthropic's Claude Code and Microsoft's GitHub Copilot to software engineers only. The shift, outlined in internal guidance updated in October and reported by Bloomberg on November 2, ends months of open experimentation in which bankers and traders across the firm were encouraged to build software with AI.

Under the revised rules, investment bankers and traders who had been using the coding assistants for daily work have lost access. Some traders had relied on the tools to monitor client trading flows, model market changes and positions, and draft written and graphic materials, according to a person familiar with the matter. The change has frustrated some of those employees, Bloomberg reported.

From vibe coding to engineers only

The reversal is striking because Goldman had publicly celebrated the experiment only months earlier. In July, chief information officer Marco Argenti praised so-called vibe coding by non-technical employees, describing how bankers and traders turned ideas into working prototypes with AI tools and then handed them to one of the firm's roughly 12,000 engineers for refinement.

Goldman spokesperson Nick Carcaterra framed the change as the natural end of a trial period. The firm makes certain AI tools available to its people to find the most effective uses across different parts of the business, he told Bloomberg, and after assessing factors including productivity and risk management, it puts them in the hands of those best suited to use them for the long term.

Who keeps access — and what is not banned

The restriction is narrower than a blanket AI ban. Software engineers keep the coding tools, and so-called strats — staff who sit at the intersection of trading, quantitative research and software engineering — are also exempt, according to the reporting. The bank continues to run its own AI systems, including the GS AI Assistant launched firmwide in June 2025 and an internal assistant that draws on models from OpenAI, Alphabet and Anthropic.

Goldman has tightened AI access before in specific settings. Earlier this year it restricted the use of Anthropic's Claude by staff in Hong Kong, a location-specific limit rather than a firmwide one. The new guidance is broader, but it still leaves general-purpose approved AI tools in place for most employees.

Wall Street's AI balancing act

The episode captures a tension spreading across finance. AI coding agents are powerful productivity tools, but they operate with their users' permissions and can touch sensitive trading data, client information and market-moving work — risks that grow when the people using them are not trained to review the code they produce.

Other banks are watching the same trade-off. Goldman is not abandoning AI; it is drawing a line around who is qualified to wield its most powerful coding form. The question for the rest of Wall Street is whether that line becomes the industry standard, or whether rivals keep betting that giving every employee a coding assistant is worth the risk.