Goldman Sachs Is Pulling Claude Code from Its Bankers and Traders

If you use AI coding assistants for financial modeling or client data analysis, your firm may restrict their use to engineers, requiring you to rely on IT for development.

Goldman Sachs has restricted the use of AI coding tools, specifically Anthropic's Claude Code and Microsoft's GitHub Copilot, to its engineering staff. Previously, traders and bankers utilized these tools for tasks such as tracking client flows and building market models. The change, detailed in updated internal guidance, follows a review of productivity and risk concerns. While quantitative researchers and software engineers, collectively known as "strats," retain access, many traders and bankers are now barred from using these agents. This move comes just weeks before Goldman Sachs is set to underwrite Anthropic's initial public offering (IPO). The bank had previously restricted Claude in Hong Kong in April, citing contract terms, but the current restriction is based on employee roles. Goldman Sachs spokesperson Nick Carcaterra stated the bank provides AI tools to staff to "determine the best use cases" across the firm.

The decision by Goldman Sachs to limit AI coding agents to engineers signals a broader industry concern regarding the governance and security of AI-generated code within regulated financial environments. While the bank previously encouraged non-technical staff to develop tools, the current stance emphasizes a shift towards centralized control and risk mitigation, especially given its role as an underwriter for Anthropic's IPO. This suggests that financial institutions are prioritizing established security protocols and code review processes over the rapid, democratized development of AI-powered tools by employees outside of dedicated engineering teams.

This restriction implies a recalibration of how banks integrate AI into their operations, moving from widespread experimentation to controlled deployment. The focus on roles and the potential for unauthorized data handling or breaches highlights the challenges of managing AI tools that can execute code. As other institutions like JPMorgan and Schwab have also tightened controls, the trend indicates that the rapid adoption of generative AI in finance may face increased scrutiny, potentially leading to a more structured and engineer-centric approach to AI development and implementation across the sector.