Goldman Sachs Pulls Claude Code and GitHub Copilot From Traders and Bankers

Goldman Sachs has reversed its open stance on AI coding tools, banning traders and bankers from using Anthropic's Claude Code and Microsoft's GitHub Copilot after an internal productivity and risk review. The decision is one of the first high-profile rollbacks of broad AI tool access inside a major Wall Street firm — and it arrives just months after Goldman's own leadership publicly celebrated the productivity gains from putting these tools in non-technical employees' hands.

Wall Street's embrace of generative AI has been rapid but uneven. Goldman, along with JPMorgan, Morgan Stanley, and Citigroup, has spent the past two years building internal AI platforms and allowing staff to experiment with commercial tools. The productivity pitch was straightforward: let analysts draft memos, coders accelerate builds, and traders automate routine scripting. That pitch just hit a wall at Goldman.

What Goldman Actually Did

According to Bloomberg Law and Finextra, Goldman has restricted Claude Code and GitHub Copilot access to approximately 12,000 engineers on its technology staff. Traders, investment bankers, and other non-engineering finance staff — a population that numbers in the tens of thousands — are now blocked from both tools.

The review that triggered the ban examined two dimensions:

• Productivity claims vs. measured output — Whether the gains reported anecdotally by non-technical staff translated into measurable business outcomes
• Risk exposure — Whether AI-assisted code written by people without formal engineering training introduced vulnerabilities, compliance gaps, or intellectual-property complications

Goldman has not publicly detailed which risk category drove the final decision, but the Finextra report notes that the IP and data-leakage concerns were prominent in the review.

The "Vibe Coding" Whiplash

The timing makes the reversal particularly notable. Goldman's Chief Information Officer had praised what the industry calls "vibe coding" — the practice of using natural-language prompts to generate working code without deep programming knowledge — as a productivity multiplier for Goldman's non-technical workforce. That endorsement was recent enough that some staff were still in the process of integrating the tools when the ban arrived.

Vibe coding has been a hot topic across the industry since early 2026. Proponents argue it democratizes software creation; critics argue it produces code that looks functional but contains subtle bugs, security holes, and licensing complications that non-engineers aren't trained to spot. Goldman's review appears to have landed firmly in the critics' camp — at least for high-stakes financial environments.

Why This Matters Beyond Goldman

Goldman's decision will reverberate across financial services for three reasons:

First, Goldman sets benchmarks. When Goldman moves on a technology practice, competitors watch. If the firm's internal review found productivity or risk problems that outweighed the gains, similar reviews at JPMorgan, Barclays, and HSBC will be harder to bury.

Second, it narrows the "democratization" narrative. The strongest case for broad AI tool access in financial firms has been that it empowers non-technical employees to build and automate. Goldman's rollback suggests that at least in regulated, high-stakes environments, unrestricted access to AI coding tools may create more liability than productivity.

Third, it has direct implications for Anthropic and Microsoft. Claude Code and GitHub Copilot both have enterprise agreements with major financial institutions. A Goldman ban — even limited to specific employee classes — puts renewal conversations at competing firms on more cautious footing.

What to Watch

Goldman has not indicated whether this is a permanent policy or a pause pending updated guardrails. Watch for whether the firm builds an internal AI coding tool with tighter data controls — similar to JPMorgan's proprietary LLM Stack — as the longer-term answer, rather than a permanent ban on the entire category.