Goldman Sachs bans AI tools for bankers and traders
November 3, 2026
Goldman Sachs bans AI tools for bankers and traders
Overview
In early 2024 Goldman Sachs issued an internal directive that limits the use of generative‑AI applications—most notably Claude, the conversational model from Anthropic—for its front‑office staff. The policy applies to bankers, traders, and other client‑facing roles. The bank cites data‑privacy, regulatory compliance, and the risk of inadvertent disclosure of proprietary information as the primary reasons for the restriction.
This article explains what the ban entails, why the firm took this step, how it affects day‑to‑day work, and what professionals in finance can do to stay productive while complying with the new rules.
Why the ban matters
Data‑security concerns
Financial institutions handle highly sensitive client data, trade secrets, and market‑moving information. Generative‑AI services are typically hosted in the cloud and require users to upload text or code snippets to the provider’s servers. Even when providers claim not to store user inputs, the possibility of accidental data capture remains. For a firm that trades billions of dollars daily, a single leak could trigger regulatory penalties, legal liability, and reputational damage.
Regulatory environment
U.S. regulators—including the SEC and the Federal Reserve—have begun issuing guidance on the use of AI in the financial sector. The guidance stresses that firms must retain control over model outputs, ensure auditability, and avoid “black‑box” decision making when it could affect client outcomes. By restricting unvetted AI tools, Goldman Sachs aligns its internal controls with emerging expectations.
Intellectual‑property protection
Traders often develop proprietary algorithms, pricing models, and execution strategies. If such code is pasted into a public AI interface, the provider could unintentionally retain it, creating a risk of intellectual‑property loss. The ban therefore protects the firm’s competitive edge.
What the policy covers
Goldman Sachs’ internal memo, which was circulated to all front‑office employees, outlines the following prohibitions:
- No use of public generative‑AI chat interfaces (e.g., Claude, ChatGPT, Bard) for drafting client communications, research notes, or internal memos.
- No submission of proprietary code to any external AI service, even for debugging or code generation.
- Prohibited use of AI‑assisted data analysis tools that send raw data to third‑party servers.
- Limited use of approved, internally hosted AI models for specific, vetted use cases (e.g., summarizing public news).
The memo also clarifies that the ban does not apply to AI tools that are fully hosted on the firm’s own infrastructure, provided they have passed the firm’s security and compliance review.
How the ban affects everyday work
Client‑facing communication
Bankers traditionally rely on email templates, pitch books, and presentation decks. The ban means they can no longer ask an AI to draft a client letter or generate a slide outline. Instead, they must use internal tools such as Microsoft Word’s built‑in editor or approved document‑automation platforms that run on the firm’s secure network.
Code development for traders
Quantitative traders often write Python scripts for data ingestion, signal generation, and order routing. The ban prevents them from copying code snippets into Claude or similar services for quick fixes. As a result, traders must rely on internal code‑review processes, shared repositories (e.g., a private GitLab instance), and internal static‑analysis tools.
Research and market analysis
Analysts frequently scan earnings calls, news releases, and macro‑economic reports. While external AI summarizers can speed up this process, the policy requires analysts to use approved internal summarization engines or perform manual note‑taking.
Collaboration
Team collaboration platforms (e.g., Slack, Teams) often integrate AI bots for drafting messages. Goldman Sachs has disabled or blocked any third‑party AI bot integrations in its corporate channels. Employees can still use the native features of the collaboration suite, but must not invoke external AI services.
Practical steps to stay productive
Even with the restrictions, there are ways to maintain efficiency without violating policy. Below are concrete actions finance professionals can take.
1. Leverage internal AI platforms
Goldman Sachs is rolling out a private‑cloud version of a large language model (LLM) that runs on its own data center. Access is granted through a secure VPN and requires multi‑factor authentication. Employees can use this model for tasks such as:
- Summarizing public earnings transcripts
- Generating boilerplate language for regulatory disclosures
- Drafting simple code snippets that do not contain proprietary logic
Because the model is hosted internally, data never leaves the firm’s network, satisfying security requirements.
2. Adopt version‑controlled code repositories
Maintain a single source of truth for all scripts and libraries. A well‑structured repository with clear branching policies reduces the need to seek external help. Features that aid productivity include:
- Pull‑request templates that enforce documentation standards
- Automated linting and unit‑test pipelines that catch errors early
- Shared utility libraries that encapsulate common patterns (e.g., data‑cleaning functions)
By centralizing code, teams can reuse proven components instead of asking an AI to “write it from scratch.”
3. Use secure document‑automation tools
Products such as Microsoft Power Automate, DocuSign, and internal workflow engines can generate routine documents (e.g., onboarding forms, compliance checklists) without external AI. These tools integrate with the firm’s identity management system, ensuring that only authorized users can trigger document creation.
4. Build a knowledge‑base of reusable assets
Create an internal wiki that stores:
- Frequently used phrasing for client communications
- Templates for pitch books and financial models
- Best‑practice guides for data handling
When a banker needs a quick draft, they can search the wiki instead of prompting an external AI.
5. Conduct regular compliance training
Understanding the policy’s boundaries is essential. Finance firms typically schedule quarterly refreshers that cover:
- What data is considered “sensitive” under the policy
- How to request an exception (e.g., for a specific research project)
- The process for reporting accidental AI usage
Employees who are clear on the rules are less likely to make inadvertent violations.
When an exception is appropriate
The policy does allow for limited exceptions. If a team identifies a legitimate business need that cannot be met with existing internal tools, they can submit a request to the firm’s AI Governance Committee. The request must include:
- A detailed description of the use case
- An assessment of data sensitivity
- A risk‑mitigation plan (e.g., anonymizing inputs before sending them to an external service)
The committee evaluates the request against regulatory guidance and the firm’s risk appetite. Approved exceptions are logged and periodically reviewed.
Broader industry context
Goldman Sachs is not alone. Other major banks—including JPMorgan Chase, Citigroup, and Bank of America—have issued similar directives limiting the use of public generative‑AI tools for front‑office staff. The common thread across