Goldman Sachs restricts AI coding tools to engineers only
Bloomberg reports a Goldman internal guideline from October limits Claude Code and GitHub Copilot to engineers, reversing its July stance praising vibe coding by non-technical staff.
Bloomberg reports that Goldman Sachs, after weighing efficiency against risk, has barred most employees from using advanced programming-oriented AI tools, drawing pushback from some staff. Bloomberg reviewed an internal guideline updated in October stating that tools such as Anthropic Claude Code and Microsoft GitHub are now restricted to engineers.
One source said the restriction frustrated some traders who had used the tools to speed up work such as monitoring client trading flows, building market volatility and position models, and generating text and chart materials. In July, Goldman chief information officer Marco Argenti had praised vibe coding by non-technical staff such as traders and bankers, describing how these employees build initial solutions with AI tools that are then refined by the firm 12,000 engineers. Goldman spokesperson Nick Carcaterra said the firm opens certain AI tools to employees to find the best application for each business line and, after a trial period, weighed efficiency and risk factors before opening them to those best placed to derive lasting value.
The latest directive does not apply to quantitative strategy staff, whose roles combine trading, quantitative research and software development. Earlier this year Goldman had separately restricted Hong Kong employees from using Anthropic Claude. Goldman has also built its own AI assistant integrating models from OpenAI, Anthropic and Alphabet, and serves as lead underwriter for Anthropic IPO.