Goldman pulled Claude Code from traders. Here’s the governance lesson.
Goldman Sachs restricted Claude Code and GitHub to engineers only. Bloomberg reported the updated internal guidance on November 2, covering a policy change made in October. Traders and bankers lost access. The bank’s “strats,” employees who sit between trading, quantitative research, and software engineering, kept it.
This happened four months after Goldman’s CIO Marco Argenti publicly praised the “vibe coding” non-technical staff were doing. Those employees had been passing their output to one of the firm’s 12,000 engineers to improve. The October guidance removed that first step.
Why Goldman drew the line where it did
Goldman spokesperson Nick Carcaterra gave a brief explanation: “After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term.” The bank has not disclosed a security incident, and it has not published the productivity data behind the call.
The strats exception is the tell. Goldman kept them because their code already flows through engineering review. The safeguard was already built in. The line was drawn at the review boundary, not the job title.
Deutsche Bank took the opposite approach
Martin Reeves, who runs engineering platforms globally for Deutsche Bank, wrote in a July post on the bank’s own site that its software developers have access to GitHub Copilot and Google’s Gemini Code Assist. Reported gains moved from 1.5 to 2.5 hours saved per week up to “material 10x gains on a number of specific tasks.”
Deutsche Bank controls the output path instead of the access list. Reeves wrote: “We define clear boundaries and never allow developers to deploy AI-generated code without review,” with developers accountable for every line implemented. Mandatory review at deployment versus a shorter license roster: two valid answers to the same governance question.
The operator takeaway
Scale this to a 40-person company and the stakes get concrete fast. Agentic coding now starts more than 40% of GitHub pull requests. Give AI coding tools to people who cannot review their own output and that review queue lands on your engineers instead of generating value.
The cost spread is wide. One analysis found a team spending $400,000 on an AI code migration while another solved the same problem for $24,000. Discipline, not the tool, explains the gap. Ungoverned access also creates an insider-risk tail: pulling access back later is harder than handing it out narrowly from the start.
Goldman also restricted Claude for staff in Hong Kong earlier this year. It runs a separate internal assistant built on models from OpenAI, Anthropic, and Alphabet. The pattern suggests access to AI coding tools is being treated as a permission tied to role, not a perk distributed broadly. That framing travels well below the scale of a bank.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
