Goldman Sachs Restricts AI Coding Tools to Its Engineers Only
· Cortex Solutions
Goldman Sachs limited Claude Code and GitHub to engineers, per Bloomberg on Nov 2, 2026. What happened, why, and what it means for AI governance.
According to a Bloomberg report published on Monday, November 2, 2026, and relayed the same day by PYMNTS and State of AI, Goldman Sachs has restricted the AI coding tools Claude Code from Anthropic and GitHub from Microsoft to its engineers only. Most of the bank's traders and bankers can no longer use these tools at work.
The news looks like an internal matter at one bank, but it raises a question every organization using AI faces today: who should be allowed to use a tool that writes code and runs it on its own?
What exactly happened?
As PYMNTS reported, citing Bloomberg, Goldman had for months allowed its bankers to use these tools, and then internal guidance issued in October limited them to engineers. State of AI says the decision followed a review of productivity and risk concerns, and that "strats", who work across trading, quantitative research and software engineering, still have access, according to a second person familiar with the matter.
According to both sources, the change frustrated some traders who had used the tools to monitor client trading flows and to create written and graphic materials. State of AI adds that they also used them to model market moves and positions.
Goldman spokesperson Nick Carcaterra told Bloomberg, as quoted by PYMNTS: "We make certain AI tools available to our people to determine the best use cases across different parts of the firm. After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term."
A retreat after enthusiasm
Notably, Goldman's chief information officer Marco Argenti praised "vibe coding" by non-technical employees as recently as July, saying traders and bankers had built tools with AI and passed them to the firm's roughly 12,000 engineers to improve.
State of AI notes that in April the bank removed Claude from its internal AI platform for bankers in Hong Kong, but that restriction was based on location and contract terms, while the new one is based on role. The bank's in-house AI assistant, which uses models from several companies, remains available firm-wide; the restriction applies to third-party agents that can write and run code by themselves.
Why a coding agent is riskier on a trading desk
State of AI explains in its analysis that a tool like Claude Code reads files, writes programs and runs them, typically with the same access as the person using it. Code an engineer writes usually goes through review, testing and deployment controls, but a small script a trader builds to watch client flows can end up in daily use without any of those. The site adds that banks must keep information barriers between groups, and a tool built by a non-specialist was not designed with those controls in mind. It also suggests cost may have played a part, since agent usage is billed by the token.
Goldman is not alone. According to PYMNTS, JPMorganChase CEO Jamie Dimon warned of "rogue agents" and said outside AI agents would not be permitted inside the bank, Bank of America placed limits on agents' access to consumer deposits, and Citigroup does not let third-party agents log into accounts or transact.
State of AI's view is that Goldman was right to restrict the tools but wrong to withdraw them with no replacement mentioned. It recommends that a named engineer take responsibility for any tool built by non-engineers before it touches production data, that existing controls apply to code an agent writes, and that companies record what people were using a tool for when it is taken away.
What it means for Saudi businesses
This is our reading at Cortex, not part of the reports cited:
- Distinguish between two kinds of tools: a chat assistant that suggests text, and an agent that writes and runs code on your machine with your permissions. The second needs a clearer policy.
- Take a simple inventory: who uses which AI tool in your organization, and on what data? Much usage happens without management knowing.
- Give every internal tool an employee builds with AI a technical owner who reviews it before it handles customer or financial data, especially given obligations under the Personal Data Protection Law.
- Don't ban and forget: record why employees were using a tool, because those are real needs that deserve an official, safe solution.
- Watch costs from day one, since open-ended use of token-billed tools can climb quickly.
The lesson from Goldman is not that AI is dangerous, but that giving access to it needs governance: who uses it, for what, and who reviews it. If you want to build AI automation that works within your organization's controls and permissions, that is what Cortex offers through its AI automation service: https://cortex.sa/services/ai-automation
Sources: Bloomberg's report (November 2, 2026) as relayed by PYMNTS and State of AI on November 2, 2026.