Goldman Sachs Restricts Employee Access to AI Coding Tools

Changes to Corporate Policy

Goldman Sachs, one of the largest players in the financial sector, is significantly tightening its policy on the use of artificial intelligence. Under new internal regulations implemented in October, most employees have lost access to advanced coding tools such as Anthropic’s Claude Code and Microsoft’s GitHub automated systems. Moving forward, only staff within the engineering department retain permission to use these tools.

Notably, Goldman Sachs traders and bankers have been actively using these technologies over the past few months. They utilized AI to monitor trading flows and generate various types of content. Reports from within the company suggest that this sharp restriction has caused dissatisfaction among some employees who had already effectively integrated AI into their workflows.

Shifting Perspectives on 'Vibe Coding'

As recently as July, Goldman Sachs Chief Information Officer (CIO) Marco Argenti publicly praised the concept of "vibe coding." This approach encouraged non-technical staff to use AI to build prototypes. The idea was straightforward: an employee, for instance from the asset management division, could create a rapid prototype with AI assistance before passing it to professional engineers to develop into a full-scale product.

It appears the company's strategy has since shifted. Goldman Sachs spokesperson Nick Carcaterra explained that the decision was driven by several factors, including the desire to increase productivity and the necessity of risk management. According to the official stance, the "experimental phase is over," and the company is now focused on ensuring these tools are used only by those with the best skills to utilise them effectively in the long term.

The Implications of the Decision

Goldman Sachs’ move highlights how major corporations are struggling to find a balance in the adoption of artificial intelligence. While AI can increase work speed, it also presents challenges regarding security, data privacy, and code quality control. In the financial sector, where the cost of error is extremely high, companies are opting for a more cautious approach.

For users in Georgia and the broader technology sector, this development serves as an important signal. It demonstrates that the democratisation of artificial intelligence in corporate environments may face temporary setbacks as organisations attempt to define the boundary between innovation and risk. It is likely that other large companies will publish similar internal guidelines in the near future, strictly regulating the use of AI tools.