Can AI Code? Why Did Goldman Sachs Take Away AI Tools from Its Bankers?

Imagine this: In the past, when you wanted to analyze complex market data, you had to ask a professional developer and wait for days. However, for the past few months, Goldman Sachs bankers and traders have been resolving such tasks simply by telling an AI, “Analyze the market flow and build a customer asset tracking model.”

But last October, Goldman Sachs put the brakes on this convenient “AI coding era.” Now, most employees, excluding 12,000 engineers, are no longer able to use AI coding tools like Claude Code or GitHub Copilot. [Source 1, Source 3, Source 4] Why did the financial giant Goldman Sachs take back the AI tools they had been using so well?

Why does this matter?

This measure means more than just a simple restriction of tools. It serves as a sign that while AI has opened an era where anyone can write code, major corporations have begun to deeply contemplate the security threats and data management responsibilities that come with “anyone being able to write code.” Particularly for a bank that handles sensitive financial information of customers, controlling where and how AI-generated code is executed, and ensuring that internally accumulated data is not leaked to external servers, is a matter directly linked to survival. [Source 6]

In simple terms: The difference between a ‘licensed professional’ and ‘recreational driving’

To use an analogy, coding is like driving a car. AI coding tools are like Advanced Driver Assistance Systems (ADAS), helping even novice drivers reach their destination safely. Until now, bankers have been “driving” their own market analysis programs using this system.

However, a driver assistance system cannot “know the road” or “obey traffic laws” for you. It is a different story if code created by AI violates the bank’s strict internal security policies or carries the risk of leaking confidential data externally. Goldman Sachs has now changed its policy so that this “driving” is once again handled only by “drivers (software engineers)” who have received professional training. [Source 6]

Current situation: Risk management ahead of productivity

In fact, as recently as July, just a few months ago, Goldman Sachs’ Chief Information Officer (CIO) Marco Argenti publicly praised the phenomenon of general employees using AI to code, calling it “vibe coding.” [Source 2]

However, the reason for the change in stance in just three months is clear. While work efficiency is important, in the financial sector where the regulatory environment is strict, “risk management” and “compliance” are the top priorities. [Source 5, Source 10] Currently, all employees within the bank, excluding the software engineer group and the “strats” group that performs quantitative analysis, are restricted from using these AI tools. [Source 1, Source 6]

Interestingly, while Goldman Sachs is taking these restrictive measures, they are simultaneously serving as a lead underwriter for the IPO of the AI technology company Anthropic. [Source 3, Source 8] This shows that Goldman Sachs is not denying the future of AI technology itself, but is in a cautious process of establishing “governance (management systems)” for how to control and utilize it in the financial field. [Source 6]

What happens next?

Goldman Sachs’ decision is expected to have a significant impact on other financial institutions as well. This is because other banks will likely continue to walk the tightrope between “improving productivity using AI” and “corporate information security.” [Source 10] It is predicted that rather than everyone freely using AI coding, we will likely see a phased introduction where use is expanded only after a “secure AI environment” that perfectly passes internal corporate security standards is established.

MindTickleBytes’ AI Reporter Perspective This case is an essential growing pain that AI technology experiences when it leaves the “laboratory” and enters the “actual business” field. It is a process where the convenience of technology is taken out of the hands of practitioners and incorporated into the management domain of the entire company. Goldman Sachs is currently checking the brakes to handle that speed, rather than the exciting speed brought by AI.

References

1. Goldman Sachs Pulls Claude Code and GitHub Copilot From Traders and Bankers. (https://aiweekly.co/alerts/goldman-sachs-pulls-claude-code-and-github-copilot-from-traders-and-bankers)
2. Goldman Sachs pulls AI coding tools from traders and bankers. (https://the-decoder.com/goldman-sachs-pulls-ai-coding-tools-from-traders-and-bankers/)
3. Goldman Sachs pulls Claude Code from most traders and bankers. (https://news.lavx.hu/article/goldman-sachs-pulls-claude-code-from-most-traders-and-bankers)
4. Goldman Sachs cuts off most staff from Claude Code and GitHub. (https://www.siliconreport.com/goldman-sachs-cuts-off-most-staff-from-claude-code-and-github)
5. Goldman Sachs pulls Claude Code and GitHub tools from traders and bankers. (https://startupfortune.com/goldman-sachs-pulls-claude-code-and-github-tools-from-traders-and-bankers/)
6. Goldman’s Claude Code Ban Exposes AI Coding Tool Governance Gap. (https://aigovernance.com/news/goldmans-claude-code-ban-exposes-ai-coding-tool-governance-gap)
7. GoldmanSachsIsPullingClaudeCodeFromItsBankersand. (https://www.thestateofai.com/news/goldman-sachs-claude-code-traders-bankers)
8. GoldmanRestrictsTradersAndBankersFrom Using Some AI Tools. (https://www.fa-mag.com/news/goldman-restricts-traders-and-bankers-from-using-some-ai-tools-89026.html)
9. HackerNews. (https://news.ycombinator.com/)