Goldman Sachs has limited access to tools such as Claude Code and GitHub to engineers, highlighting Wall Street’s growing challenge of balancing AI experimentation with productivity, cost, and risk controls.
Goldman Sachs Group Inc. has restricted most employees from using certain advanced artificial intelligence coding tools after reviewing their productivity and risk implications, according to a Bloomberg report published November 2, 2026.
The bank had previously allowed bankers and traders to experiment with tools, including Anthropic’s Claude Code and Microsoft’s GitHub. However, internal guidance updated in October now limits their use to engineers, Bloomberg reported.
The change has frustrated some traders who had used the tools to accelerate tasks such as monitoring client trading flows, modeling market changes and positions, and creating written or graphic content.
The decision reflects a wider shift across financial services. Banks are investing heavily in AI while tightening controls over how employees use increasingly capable tools.
Goldman Moves From Broad Experimentation to Controlled Access
Goldman’s updated policy follows a period of experimentation with AI coding tools across different parts of the firm.
In July, Chief Information Officer Marco Argenti discussed how non-technical employees—including traders and bankers—were using “vibe coding” to develop solutions. Those employees could then pass their creations to Goldman’s engineering teams for further development.
The bank’s latest guidance appears to narrow that model by reserving certain tools for employees with engineering roles.
Goldman spokesperson Nick Carcaterra said the firm makes AI tools available to employees to identify useful applications across the business. After an experimentation period, the bank assesses factors including productivity and risk management before deciding who is best suited to use the tools over the long term.
The restriction does not apply to “strats”—employees who work across trading, quantitative research, and software engineering—according to a person familiar with the matter cited by Bloomberg.
The distinction suggests Goldman is not abandoning AI-assisted development. Instead, it is drawing boundaries around who can use specific tools and under what conditions.
Why AI Coding Tools Matter to Traders and Bankers
Advanced coding assistants can help employees build small applications, automate repetitive tasks, and explore data without writing every line of code themselves.
For traders, these tools can support workflows involving client activity, market scenarios, and position analysis. Bankers may use them to create content, automate internal processes, or build tools for recurring tasks.
However, the ability to generate code quickly does not automatically make a tool suitable for use in a regulated financial institution.
AI-generated applications may require testing, security review, access controls, and ongoing maintenance. They can also create operational risks if they interact with sensitive data or systems without adequate oversight.
Goldman’s decision therefore highlights a tension: tools that make employees more productive can also expand the number of applications being built outside traditional engineering processes.
The bank’s approach appears to place greater emphasis on controlled deployment after experimentation, rather than unrestricted access for every employee.
Wall Street Weighs Productivity Against AI Costs and Risk
Goldman’s move comes as financial institutions work to define how thousands of employees should use AI models.
Banks are spending millions of dollars on AI tools while also assessing cybersecurity risks, costs, and productivity outcomes. The challenge is not simply whether to adopt AI, but how to match the right model and tool to each task.
JPMorgan Chase said in July that it was seeking to help employees determine which model to use for specific purposes. That approach aims to avoid using expensive models for simpler tasks, such as summarizing an analyst report.
Goldman has also developed an AI assistant that combines models from OpenAI, Anthropic, and Alphabet. The bank’s wider AI strategy therefore continues, even as access to particular coding tools becomes more restricted.
The latest decision also follows an earlier, location-specific restriction on Anthropic’s Claude for Goldman employees in Hong Kong.
Together, these developments show that AI access policies can vary by tool, employee role, and location.
A Complex Relationship With AI Providers
Goldman’s AI strategy is unfolding alongside its business relationships with major technology companies.
The bank has been named lead underwriter for Anthropic’s planned initial public offering, according to Bloomberg, and has partnered with Nvidia to arrange major debt deals.
At the same time, Goldman is evaluating how AI tools should be used internally. These commercial relationships do not establish the reason for the restrictions, but they illustrate how AI companies are becoming both technology providers and important participants in financial markets.
In May, Goldman President John Waldron described the firm as a “human assembly line” with significant potential for automation. The bank’s latest move shows that automation is not necessarily synonymous with giving every employee unrestricted access to every AI tool.
Instead, the firm appears to be pursuing a more selective approach, combining experimentation with role-based access and risk review.
What Goldman’s Decision Signals for Financial Services
Goldman’s restrictions offer a snapshot of the next phase of enterprise AI adoption in finance.
Early experimentation showed that employees outside engineering could use AI to build useful tools. The next challenge is determining which of those tools can be deployed safely, maintained effectively, and scaled across a complex institution.
For financial firms, that may mean separating experimentation from production use. Employees could identify potential applications, while engineering and control teams assess security, reliability, and operational impact before wider deployment.
The approach may also encourage banks to develop internal AI assistants and approved toolsets that give employees access to useful capabilities within defined boundaries.
Goldman has not disclosed detailed productivity measurements or the specific risk findings behind its latest restrictions. The announcement therefore does not establish whether the policy will increase or reduce overall productivity.
The broader lesson is that financial institutions are moving beyond the question of whether employees can use AI. They are now deciding who should use which tools, for what tasks, and under what controls.


Comments
No comments yet. Be the first to share your thoughts.
Leave a comment