Goldman Sachs restricts AI tools to engineering teams

The shift in access

Goldman Sachs is tightening the reins on how its employees interact with artificial intelligence. According to a report by Bloomberg News, the firm is moving to restrict several specific AI tools to its engineering teams only. This change follows a period where the bank allowed a wider range of staff to experiment with these technologies.

The tools being restricted include Microsoft’s GitHub and Anthropic’s Claude Code. These platforms are often used to write, test, and manage software code. By limiting them to engineers, the firm is centralizing the use of high-powered development tools within a specific technical department.

A spokesperson for Goldman Sachs, Nick Carcaterra, explained that the firm is evaluating various factors. He stated that after a period of experimentation, the bank is putting tools in the hands of those best suited to leverage them for the long term. This suggests a move toward a more controlled, risk-managed environment for AI deployment.

From 'vibe coding' to oversight

Before this restriction, non-technical employees at the bank were engaging in what Chief Information Officer Marco Argenti called 'vibe coding.' This term refers to non-technical workers using AI to develop software solutions based on their general ideas or 'vibes' rather than formal programming skills. These workers would then pass their creations to the firm's 12,000 engineers to enhance and finalize.

Bankers had been using these AI tools for a variety of practical tasks in their daily workflows. Some traders used the technology to monitor client trading flows more efficiently. Others utilized the tools to generate written content and graphic materials for the firm's clients.

The move to restrict these tools has reportedly caused some frustration among the trading desks. While the bank wants to manage risk, the traders who had integrated these tools into their daily routines now face limitations. The firm is balancing the desire for productivity with the need to manage cybersecurity and operational risks.

The broader banking landscape

Goldman Sachs is not alone in navigating the complexities of AI integration. Many large financial institutions are spending millions of dollars on AI tools while simultaneously grappling with concerns over costs and security. The industry is currently trying to determine the safest way to deploy these technologies without exposing sensitive data.

JPMorganChase CEO Jamie Dimon recently warned about the dangers of 'rogue agents.' During a third-quarter earnings call, he stated that outside AI agents would not be permitted inside the bank. This reflects a broader trend of financial institutions creating 'walls' around their internal systems to prevent unauthorized AI access.

Other banks are taking different approaches to these restrictions. Bank of America has implemented conditions rather than a total ban, specifically limiting AI agents' access to consumer deposits. They are currently developing protocols to decide who is liable when an AI authorizes a payment. Citigroup has also drawn a similar line, refusing to let third-party agents log into accounts or perform transactions.

Labor and productivity trends

While large banks are focusing on security, research into the broader economy shows significant shifts in how AI affects the workforce. Recent PYMNTS Intelligence research found that 60% of small and medium-sized businesses (SMBs) using AI reported that the technology allows them to produce more output without hiring new employees.

The research suggests that the most significant impact of AI on the labor market might not be the elimination of existing jobs. Instead, it may be that companies no longer feel the need to create new positions to handle routine tasks. As automation makes standard work cheaper, the demand for new labor positions may decrease.

This shift could create a different kind of market for human workers. The research indicates that as routine work becomes automated, there may be a growing need for human judgment and accountability. These are qualities that remain much harder for current artificial intelligence systems to replicate or automate effectively.

What comes next for AI

The move by Goldman Sachs signals a transition from a period of 'open' experimentation to one of structured corporate governance. The bank is moving toward a model where AI is a tool for specialized experts rather than a general-purpose utility for every employee. This allows the firm to monitor exactly how data is being processed and who is responsible for the outputs.

For the banking industry, the next phase will likely involve the development of standardized safety protocols. Banks need to find a way to harness the productivity gains of AI while satisfying regulators and protecting against cybersecurity threats. We can expect to see more specific rules regarding which types of AI tasks are 'safe' for general staff versus which require engineering oversight.

As the technology matures, the focus will likely shift from what the AI can do to how it can be done safely. The tension between the 'vibe coding' era of rapid, unguided creation and the new era of restricted, high-security deployment will define the next year of corporate AI strategy. The goal for firms is to maintain a competitive edge without creating new vulnerabilities.