Goldman Sachs Restricts AI Coding Tools to Engineering Staff

Goldman Sachs has updated its internal guidance to restrict the use of AI coding agents, specifically Anthropic's Claude Code and Microsoft's GitHub, to the bank's engineers, as State of AI first reported. According to Bloomberg News, the move follows a review focused on productivity and risk concerns.

Previously, non-technical employees used these tools to build market models, track client flows, and create graphic materials. In July, Chief Information Officer Marco Argenti praised this "vibe coding" approach, noting that traders and bankers could build initial versions of tools to be later refined by the firm's approximately 12,000 engineers. However, the October guidance removes this access for most traders and bankers, though "strats" working in quantitative research, software engineering, and trading maintain access.

Risk factors driving the decision include the lack of oversight for scripts built by non-engineers, which may bypass review, testing, and deployment controls. Additionally, there are concerns regarding information barriers between deal-making bankers and traders, as well as the potential for financial losses if a trader uses code they cannot fully understand to model positions.

This shift occurs as Goldman Sachs and Morgan Stanley serve as lead underwriters for Anthropic's IPO, which is expected to list on Nasdaq under the ticker ANT by the end of November. Goldman previously collaborated with Anthropic engineers to build agents for client onboarding and trade accounting, which remain managed by engineering teams.

While Goldman is tightening access, other firms are taking different paths. Barclays announced in October that it expects half of its developers to use Claude Code by year-end, coinciding with the announcement of 5,000 job cuts that include developers.