New York: Goldman Sachs has restricted access to advanced artificial intelligence (AI) coding tools, including Anthropic’s Claude Code and Microsoft’s GitHub, for most employees, limiting their use primarily to software engineers. The decision has frustrated some workers, particularly traders who had been using the applications to speed up routine tasks and develop tools for their daily work, Bloomberg reported.
The move marks a shift in the bank’s approach to workplace AI adoption. Goldman Sachs had previously encouraged employees across trading and banking teams to experiment with AI-powered applications, allowing them to explore how the technology could improve productivity, analyse information and simplify workflows.
However, following months of experimentation, the bank has reassessed how these tools should be deployed. Productivity, risk management and the suitability of different applications for particular roles were among the factors considered before access was tightened.
The restrictions reflect a broader challenge facing major financial institutions: how to benefit from rapidly developing AI technology without exposing sensitive operations to additional cybersecurity, regulatory and operational risks.
Goldman Sachs narrows access after AI experiments
Employees across Goldman Sachs had been using advanced coding tools for a variety of tasks, including monitoring client trading patterns, analysing market movements, preparing reports and creating visual content.
These applications offered employees an opportunity to automate certain activities and build simple tools without relying entirely on dedicated software development teams.
For traders and banking professionals, the ability to use AI coding assistants meant they could experiment with applications suited to their specific responsibilities. Instead of waiting for a software engineer to develop every small tool, employees with limited technical expertise could create early versions themselves.
The latest decision changes that approach. Access to tools such as Claude Code and GitHub has been restricted for most employees, with software engineers being given priority.
The change has reportedly disappointed some traders who had incorporated these applications into their working routines. Their frustration reflects the growing dependence of financial professionals on AI tools for tasks that previously required more manual effort or assistance from technology teams.
The restrictions do not mean Goldman Sachs has abandoned AI adoption. Instead, the bank appears to be narrowing access to ensure that advanced coding applications are used by employees considered best equipped to manage them over the longer term.
Earlier push encouraged ‘vibe coding’
The decision is notable because it follows an earlier effort by Goldman Sachs to encourage wider experimentation with AI.
In July, the bank’s Chief Information Officer, Marco Argenti, praised the concept of ‘vibe coding’. The term describes the use of AI tools to create software and basic applications through natural-language instructions, allowing people without formal programming expertise to participate in software development.
Under this approach, bankers and traders could build preliminary versions of AI-powered tools based on their day-to-day requirements. These early applications could subsequently be handed over to the bank’s engineering teams for further development, testing and refinement.
Goldman Sachs has around 12,000 engineers who can support the development of such tools. The earlier approach was intended to make it easier for employees across different departments to identify useful applications before engineers worked on more advanced versions.
For example, a trader familiar with a particular market workflow could use an AI coding assistant to create a basic tool addressing a specific requirement. The engineering team could then examine the application and determine whether it was suitable for wider deployment.
The latest restrictions indicate that the bank is moving towards a more controlled model after assessing the results of its experimentation.
Bank cites productivity and risk management
Goldman Sachs has defended its decision by explaining that access to AI applications is determined after evaluating their usefulness and potential risks.
A spokesperson for the bank said: “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.”
The statement suggests that the bank is not treating access to advanced AI tools as a benefit that should automatically extend to every employee.
Instead, it wants to identify which roles can make the most effective use of particular applications while ensuring that their deployment remains manageable.
For a financial institution, the potential risks associated with AI coding tools can extend beyond the quality of the software they generate. Applications developed by employees may interact with internal systems, handle sensitive information or influence workflows connected to financial operations.
These considerations make it important for banks to maintain oversight over how software is created, tested and used.
The bank’s decision also highlights the difference between experimenting with AI and allowing applications generated through AI tools to become part of regular business operations. While experimentation can help identify new efficiencies, wider deployment requires appropriate technical checks and risk controls.
Specialist ‘strats’ retain certain exemptions
The restrictions are not universal across Goldman Sachs.
Certain specialist employees, known as ‘strats’, remain outside the scope of the latest limitations and continue to enjoy exemptions.
Strats are professionals whose responsibilities combine trading, mathematical research and software development. Their work often involves applying quantitative methods and technical expertise to financial markets and trading-related activities.
Because their roles already span areas that require programming, mathematical analysis and financial knowledge, they have different requirements from employees whose primary responsibilities are not technical.
Their continued access highlights the difficulty of designing a single AI policy for an organisation with diverse job functions.
A trader who primarily focuses on market activity may use an AI coding assistant differently from a quantitative specialist who develops mathematical models and software systems. Similarly, software engineers may require advanced coding tools as part of their regular responsibilities.
Goldman Sachs’ exemptions suggest that the bank is taking these differences into account when deciding who can use the applications.
The distinction also indicates that the restrictions are aimed at controlling access rather than eliminating the use of advanced AI coding tools across the organisation.
Wall Street banks reassess AI spending
Goldman Sachs is not alone in trying to determine how extensively employees should use AI applications.
Major financial institutions are investing in AI technology while examining whether the benefits justify the associated costs and operational risks.
Earlier this year, JPMorgan Chase & Co urged employees to match AI models to the complexity of their assignments. The guidance encouraged workers to avoid using expensive and sophisticated models for routine tasks that could be handled by simpler alternatives.
The approach reflects a growing focus on the cost of running AI systems. Different models and applications can have different computing requirements and expenses, making it important for companies to select tools according to the nature of the work.
For large banks, widespread AI adoption also raises questions about data protection, cybersecurity, reliability and the possibility of errors in automated processes.
Financial institutions handle sensitive client information and conduct activities that can have significant financial and regulatory consequences. As a result, they cannot evaluate AI applications solely on the basis of how quickly they complete a task.
The technology must also meet internal standards for security, accuracy and operational reliability.
Goldman Sachs’ latest move fits into this wider effort to balance employee access with institutional oversight.
Goldman Sachs continues investing in AI
Despite tightening access to certain coding tools, Goldman Sachs continues to develop its broader AI capabilities.
The bank has built an in-house AI assistant that draws on models developed by technology companies including OpenAI, Anthropic and Alphabet.
Such an approach allows the institution to use capabilities developed by external technology providers while building an internal platform suited to its own requirements.
The bank’s wider business activities also reflect its interest in the growth of the technology sector. Goldman Sachs has recently served as a lead underwriter for major technology listings in the pipeline and has worked with Nvidia Corporation on large-scale debt transactions.
These activities highlight the distinction between Goldman Sachs’ confidence in the commercial potential of AI and its approach to allowing employees to use individual AI applications.
The bank can continue investing in AI infrastructure, supporting technology companies and developing internal assistants while restricting access to certain tools for employees whose roles may not require them.
Balancing innovation with control
The decision to limit access to Claude Code and GitHub illustrates a broader shift in how financial institutions are approaching workplace AI.
The initial phase of adoption encouraged employees to experiment with applications that could automate routine work and improve productivity. The next phase is increasingly focused on identifying which tools should be used, who should have access and how their deployment can be monitored.
For Goldman Sachs, the challenge is to retain the benefits of employee-led experimentation without creating unnecessary operational or security risks.
The continued exemptions for specialist ‘strats’ and the priority given to software engineers show that the bank is not applying a blanket ban. Instead, it is differentiating access according to employees’ responsibilities and technical requirements.
As AI coding assistants become more capable, similar decisions are likely to remain important for financial institutions seeking to integrate the technology into their operations.
Goldman Sachs’ latest restrictions underline a central question for Wall Street: how can banks encourage innovation while ensuring that increasingly powerful AI tools are used in a controlled and accountable manner?