The Zhitong Finance App learned that, according to reports, Goldman Sachs (GS.US) has banned some employees from using advanced artificial intelligence (AI) tools designed specifically for programming after evaluating productivity and risk concerns. This means that after several months of actively promoting AI applications, the Wall Street giant has begun to redefine the scope of use of some AI tools.

The report said that for the previous few months, Goldman Sachs bankers had been using tools related to Anthropic and Microsoft (MSFT.US) GitHub in their work. However, according to the bank's internal guidelines updated in October, the right to use these tools is currently restricted to the group of engineers.

This change has already sparked dissatisfaction among some employees. People familiar with the matter revealed that some traders have previously used these AI tools to speed up tasks including monitoring customer transaction flows, modeling market changes and positions, and creating text and graphic content, so they are frustrated by the sudden tightening of authority on related tools.

Notably, this approach clearly contrasts with Goldman Sachs's positive attitude towards AI in July of this year. At the time, Goldman Sachs Chief Information Officer Marco Argenti publicly discussed the so-called “vibe coding,” where non-technical employees use AI tools to develop their own software or solutions. According to the model at the time, employees most familiar with business requirements, such as traders and bankers, can first use AI to quickly build solutions, and then hand over these results to one of Goldman Sachs's 12,000 engineers for refinement and actual use.

The advantage of this model is that it can lower the threshold for non-technical employees to develop software while providing engineering teams with more clear business requirements. For traders and bankers, they can first use AI to build a prototype without waiting for a schedule from the engineering department, so they can quickly verify whether an idea has real value.

However, as the capabilities of AI coding agents continue to improve, this model also poses new risks: AI-generated code may be directly used to process customer information, transaction data, or internal systems, and users themselves may not have sufficient software engineering capabilities to identify security flaws, logic errors, or data permission issues in the code.

This risk is particularly sensitive for a heavily regulated global bank. Generative AI may not only generate wrong answers, but may also involve issues such as sensitive data breaches, cybersecurity, intellectual property rights, model output reliability, and regulatory records. As a result, financial institutions need to find a balance between increasing employee productivity and maintaining risk control.

Goldman Sachs's latest policy actually reflects a larger change in Wall Street's AI applications. Over the past two years, financial institutions have generally viewed generative AI as an important tool to improve employee productivity. With the rapid improvement of the capabilities of Claude, ChatGPT, and various AI coding agents, AI has gradually evolved from a simple text generation tool to an “intelligent agent” that can write code, analyze data, automate tasks, and even build complete workflows.

However, the stronger AI capabilities, the more complex the governance problems faced by enterprises. Reports have recently indicated that the use of unapproved generative AI tools by employees within companies is creating a so-called “shadow AI” (shadow AI) problem. Key risks faced by enterprises include loss of control over confidential data, cybersecurity incidents, model output errors or bias, intellectual property issues, and lack of reliable records of important decisions.

Therefore, instead of simply prohibiting employees from using AI, more and more enterprises are adopting more detailed governance methods, such as classifying different AI usage scenarios based on data sensitivity, error consequences, and the importance of business decisions, and setting different permissions for different employee groups. Seen from this perspective, Goldman Sachs currently mainly hands over advanced programming tools such as Claude Code and GitHub to engineers, which can also be seen as a shift in its AI governance model from “open experiments” to “specialized management.”

Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.