Goldman Sachs restricts AI coding tools to engineers

Goldman Sachs has restricted most of its employees from using advanced artificial intelligence coding tools, according to internal guidelines updated in October (the-decoder.com). The policy change cuts off professionals outside of engineering, such as traders and bankers, from platforms including Anthropic's Claude Code and Microsoft's GitHub (the-decoder.com, cryptobriefing.com). Goldman's quantitative research, trading, and software engineering roles known as "strats" are excluded from the restriction (cryptobriefing.com).

What happened

For months, non-technical personnel across trading and asset management utilized AI coding systems to build prototypes, monitor client trading flows, model market conditions, and produce written and graphic content (the-decoder.com, cryptobriefing.com). The rollback follows earlier praise from Goldman Chief Information Officer Marco Argenti, who in July highlighted a "vibe coding" workflow where non-technical employees rapidly prototyped ideas with AI before handing them to engineers for production scaling (the-decoder.com, cryptobriefing.com). That broad experimentation phase has now concluded (the-decoder.com, cryptobriefing.com).

Why it matters

The restriction highlights the tension between adopting AI to drive efficiency and protecting sensitive institutional data (note.com). Financial institutions operate under strict confidentiality requirements, meaning that AI-driven code generation and data processing can introduce security vulnerabilities or risks of leaking confidential transaction flows (note.com).

Context

Goldman spokesperson Nick Carcaterra stated that the decision was driven by a "variety of factors including productivity and risk management" (the-decoder.com). Carcaterra added that the firm is now limiting the tools to the personnel "who are best suited to leverage them for the long term" (the-decoder.com). The pivot has reportedly generated internal frustration among employees who had integrated the technology into their daily workflows (the-decoder.com, cryptobriefing.com).