Goldman Sachs Pulls AI Coding Tools From Traders and Bankers

Goldman Sachs has restricted most traders and bankers from using advanced AI coding tools, including Anthropic’s Claude Code and Microsoft’s GitHub, following an internal review of productivity and risk management. The decision reverses an earlier approach in which non-technical employees were encouraged to experiment with AI-assisted coding, build prototypes and pass them to engineers for further development.

According to Bloomberg, internal guidance updated in October now limits these tools primarily to engineering teams. The restriction does not cover specialised employees known as “strats”, who work across trading, quantitative research and software engineering. The move highlights the challenges financial institutions face as they attempt to introduce powerful AI tools while maintaining strict controls over sensitive data, technology systems and business operations.

Key Takeaways

• Goldman Sachs has restricted most traders and bankers from using advanced AI coding tools.
• The restrictions affect Anthropic’s Claude Code and Microsoft’s GitHub.
• Updated internal guidance from October limits access primarily to engineers.
• Some traders had used the tools to monitor client trading flows, model market movements and create written and visual content.
• Goldman cites productivity and risk management as factors behind the decision.
• Specialised employees known as “strats” are reportedly exempt.
• The decision reverses an earlier encouragement of AI-assisted development among non-technical employees.
• Goldman continues to develop its own AI assistant using models from OpenAI, Anthropic and Google.

What Has Goldman Sachs Changed?

Goldman Sachs had previously allowed employees outside its engineering department to experiment with advanced AI coding tools.

These included Claude Code, Anthropic’s AI-powered coding agent, and Microsoft’s GitHub tools.

The applications allowed employees to generate code, develop small software solutions and automate certain work processes without necessarily having traditional programming expertise.

However, the bank has now narrowed access.

Bloomberg reported that internal guidance updated in October limits the use of these advanced coding tools primarily to engineers.

The restriction affects traders and bankers who had previously been able to experiment with the software as part of their daily work.

Goldman spokesperson Nick Carcaterra explained that the bank makes AI tools available to identify useful applications across different departments. After an experimentation period, the company evaluates productivity, risk management and other considerations before deciding which teams should retain access.

The bank described the decision as part of its process of determining which employees are best positioned to use particular tools over the long term.

Importantly, this is not a complete ban on artificial intelligence at Goldman Sachs. The restriction concerns certain advanced coding tools, rather than every AI application used by employees.

Why Were Traders and Bankers Using AI Coding Tools?

The tools had become useful to employees who wanted to automate repetitive tasks and develop customised solutions for their work.

According to Bloomberg, some traders were using AI coding assistants for several activities.

1. Monitoring Client Trading Flows

Traders need to monitor activity across financial markets and understand how clients are positioning themselves.

AI-assisted coding tools allowed some employees to develop applications that helped track client trading flows and organise relevant information.

Such tools could potentially reduce the time spent collecting and analysing information from different sources.

2. Modelling Market Movements

Financial markets change rapidly, and traders frequently need to examine how different market scenarios could affect their positions.

Some employees used AI tools to build models and applications that helped them analyse market movements and potential portfolio exposures.

These prototypes could be developed without waiting for a complete software development project.

3. Creating Reports and Visual Content

AI coding assistants were also used to generate written and graphical material.

Employees could create customised dashboards, visualisations and internal reports to communicate information more efficiently.

For non-technical employees, these tools offered a way to turn an idea into a working prototype using natural-language instructions.

The reported restrictions could therefore affect employees who had incorporated AI-assisted development into their everyday workflows.

Goldman Previously Encouraged Vibe Coding

The latest decision represents a notable change from Goldman’s earlier enthusiasm for AI-assisted software development.

In July 2026, Goldman Sachs Chief Information Officer Marco Argenti publicly discussed the benefits of what is commonly called vibe coding.

Vibe coding refers to using natural-language instructions to ask an AI system to generate software, rather than writing every line of code manually.

Argenti described how employees outside traditional technology roles, including traders and bankers, could use AI tools to develop prototypes based on their ideas.

These prototypes could then be handed over to professional engineers for further development, testing and deployment.

Goldman has approximately 12,000 engineers, according to Argenti’s earlier comments.

The approach offered a way to involve more employees in software development without expecting everyone to become a professional programmer.

However, the latest restrictions suggest the bank has decided that unrestricted access to advanced coding agents is not the right long-term arrangement for every department.

Why Is Goldman Sachs Concerned About AI Coding Tools?

Financial institutions operate under strict requirements involving data security, regulatory compliance, operational reliability and internal controls.

AI coding agents introduce additional considerations because they can do more than generate text.

Depending on their permissions, these tools can inspect codebases, modify files, execute commands and interact with development environments.

That creates potential risks when employees use them without sufficient technical supervision.

Cybersecurity risks: AI agents may interact with sensitive files, internal systems or code repositories. Incorrect permissions or vulnerabilities could expose confidential information.

Code quality: AI-generated code can contain errors, security weaknesses or unexpected behaviour.

Regulatory requirements: Software used in financial operations must meet internal governance and compliance requirements.

Operational risks: An improperly tested application could produce inaccurate calculations or affect important business workflows.

Productivity concerns: The time saved through AI-generated code may be offset by debugging, testing, reviewing and maintaining the resulting applications.

Goldman has not identified one specific security incident or coding failure as the reason for the restrictions. Its spokesperson referred more broadly to productivity and risk management.

Why Are Specialised Employees Called Strats Exempt?

The reported restriction does not apply to employees known as strats.

At Goldman Sachs, strats work at the intersection of quantitative research, trading and software engineering.

They typically combine mathematical, financial and programming expertise to develop models and technology used by trading teams.

Their responsibilities can include pricing financial products, analysing risk, building quantitative models and developing specialised trading systems.

Because their roles already involve substantial technical work, the bank appears to consider them better equipped to use advanced coding tools within controlled development environments.

Their exemption illustrates that Goldman’s decision is not necessarily about eliminating AI-assisted coding from financial operations.

Instead, the bank appears to be narrowing access towards employees with the technical expertise to manage the associated risks.

Goldman Sachs Is Still Investing in Artificial Intelligence

Despite restricting access to certain tools, Goldman Sachs continues to expand its AI initiatives.

The bank has developed an internal AI assistant that combines models from several technology companies, including OpenAI, Anthropic and Google.

This approach allows Goldman to use different AI models for different tasks rather than depending on a single provider.

The company is also exploring ways to automate repetitive business processes and improve employee productivity.

Its strategy reflects a broader trend across financial services: banks want to benefit from AI while maintaining control over how the technology interacts with their systems and data.

Goldman is therefore not abandoning AI adoption. It is adjusting which employees can use particular categories of AI software and under what conditions.

Other Wall Street Banks Are Also Tightening AI Controls

Goldman Sachs is not the only financial institution reconsidering how employees should use increasingly capable AI systems.

JPMorgan Chase has been encouraging employees to match AI models to the tasks they perform.

The objective is partly to manage costs by ensuring employees do not use expensive models for relatively simple activities, such as summarising analyst reports.

Other banks are also establishing restrictions around external AI agents.

JPMorgan CEO Jamie Dimon has raised concerns about unauthorised or poorly controlled AI agents accessing financial systems.

Bank of America and Citigroup have also introduced limits on third-party AI agents interacting with customer accounts and carrying out financial transactions.

These developments demonstrate that financial institutions are moving towards more structured AI governance rather than allowing unrestricted experimentation.

The Bigger Picture

Goldman’s decision highlights an important distinction in enterprise AI adoption: making AI available to employees is not the same as allowing every employee to deploy AI-generated software in a production environment.

Vibe coding has lowered the barrier to creating software. Employees who previously depended on engineering teams can now build prototypes using natural-language instructions.

However, financial institutions operate in environments where software errors can have significant financial, regulatory and operational consequences.

The emerging model may involve giving employees access to AI for experimentation while reserving advanced coding agents and production deployment for specialised technical teams.

For the broader technology industry, this creates a new question about the future of AI-assisted work: how can companies encourage employees to innovate without sacrificing security, reliability and accountability?

Looking Ahead

Goldman Sachs will continue to determine which AI tools provide measurable productivity benefits while meeting its internal risk requirements. The latest restrictions suggest that advanced coding agents will be treated more like specialised engineering tools than general-purpose applications available to every employee.

For the financial industry, the next phase of AI adoption is likely to focus on governance, access permissions, testing and accountability. Banks are not necessarily stepping away from AI; they are working out how to deploy increasingly capable systems without introducing unacceptable risks into critical financial operations.