# Goldman Sachs Restricts AI Coding Tools for Bankers and Traders Bank limits Claude Code and GitHub access to engineers amid risk and productivity concerns. - **Source:** [Hacker News (news.ycombinator.com)](https://brifing.si/source/13c923cd-6d99-442b-9b8b-1181f4d8cd91) - **Published:** 2026-11-08 - **Topic:** [Finance](https://brifing.si/c/finance) - **Original article:** - **On Brifing:** https://brifing.si/s/8e961588-cd50-44f7-80b6-f1fef0e50b6d ## Policy Shift on AI Goldman Sachs has restricted the use of Anthropic’s Claude Code and Microsoft’s GitHub to its engineering staff. Previously, traders and bankers utilized these AI tools to monitor client flows and build market models. The updated October guidance aims to mitigate risks associated with unreviewed code running on sensitive financial systems. ## Engineering-Only Access Model The bank currently employs approximately 12,000 engineers out of a total workforce of 47,000. Under the new policy, only these technical staff members retain access to third-party coding agents. This decision reverses earlier initiatives that encouraged non-technical employees to build their own software tools to improve daily operational efficiency. ## Information Barrier Risks Financial institutions must maintain strict information barriers between departments, such as deal-making bankers and market traders. AI-generated code often lacks these built-in regulatory controls, creating potential compliance vulnerabilities. Because traders may not fully understand the code they deploy, the bank faces significant risks regarding data security and regulatory oversight. ## Shift from Vibe Coding As recently as July, CIO Marco Argenti promoted 'vibe coding,' where non-technical staff built tools for engineers to refine. This approach allowed rapid prototyping but bypassed standard testing and deployment protocols. The new restriction effectively ends this experiment, centralizing software development responsibilities back within the firm's dedicated engineering departments. ## IPO Underwriting Conflict Goldman Sachs is currently a lead underwriter for Anthropic’s upcoming IPO, expected to list on Nasdaq before the end of November. Ironically, the bank’s bankers are now tasked with marketing shares of a company whose primary coding product they are no longer permitted to use for their own daily workflows. ## Operational and Cost Factors Productivity reviews and token-based billing costs likely influenced the bank’s decision to limit access. Similar to other firms like Accenture, Goldman is managing the financial impact of widespread AI agent usage. By restricting tools to engineers, the bank can better control software quality while managing the rising costs of AI. ## Regional Compliance Challenges Goldman previously removed Claude from its Hong Kong internal platform in April due to strict contract interpretations. JPMorgan also restricted Claude in the same region during June. These actions highlight how global banks are navigating fragmented regulatory environments and varying third-party software support while deploying advanced AI coding agents. ## Future Governance Requirements The situation underscores the need for clear ownership of agent-built code within regulated industries. Future AI adoption requires that a designated engineer takes responsibility for any tool before it interacts with production data. Organizations must balance the speed of AI-driven innovation with the necessity of maintaining rigorous software approval standards. ## Sources covering this story - https://www.thestateofai.com/news/goldman-sachs-claude-code-traders-bankers — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._