Goldman AI coding tools are now limited to engineers. Goldman Sachs has restricted Anthropic’s Claude Code and Microsoft’s GitHub after a review of productivity and risk, ending months of broader access for bankers and traders, according to October guidance reported by Bloomberg on Nov. 2. The pullback matters because those tools had already entered trading workflows, not only software teams.
Key Takeaways
- Goldman limits Claude Code and GitHub to engineers after a productivity and risk review.
- Some traders who monitored flows, modeled positions, and made content are frustrated.
- Strats stay exempt; Hong Kong employees were restricted from Claude earlier this year.
- Dimon said outside AI agents would not be permitted inside JPMorgan.
What the Goldman AI Coding Tools Limit Changes
For months, Goldman let bankers use Claude Code and GitHub on the job, PYMNTS reported, citing Bloomberg. Guidance published last month now confines those tools to engineers. Crypto Briefing reported that a spokesperson said the bank had concluded an experimentation period and was placing the tools with employees best suited to use them long term.
The limit is role-based, not a firmwide exit from coding assistance:
- Engineers keep access under the October guidance.
- Bankers and traders, other than exempt strats, lose the broader access they had.
- Strats, whose roles combine trading, quantitative research, and software engineering, stay exempt.
- Earlier this year, the bank restricted Hong Kong-based employees from using Anthropic’s Claude, Odaily reported.
Why Some Traders Object to the Pullback
Some traders are frustrated, a source familiar with the matter told Bloomberg. They had used the tools to monitor client trading flows, model market conditions and positions, and create written and graphic material. Odaily reported that traders had used the tools to speed up that work, and that the directive sparked discontent among some staff.
In analysis terms, those uses are a harder control problem than an internal prototype, because they touch client flows, positions, and material that can leave the desk.
From Vibe Coding to a Narrower Mandate
The new line contrasts with July comments from Chief Information Officer Marco Argenti. He touted “vibe coding” by non-technical employees who built solutions with AI tools and passed them to one of the firm’s 12,000 engineers to enhance, PYMNTS reported.
“We make certain AI tools available to our people to determine the best use cases across different parts of the firm,” Nick Carcaterra, a Goldman spokesperson, told Bloomberg. “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.”
Read as operating policy, the quote describes a sequence: make tools available, assess productivity and risk, then narrow who keeps them.
How Peer Banks Are Limiting AI Action Rights
PYMNTS noted the Goldman move as banks spend millions on AI while trying to determine how to deal with rising concerns related to cost, productivity, and cybersecurity risk. The same account described separate limits on external agents.
| Institution | Restriction reported |
|---|---|
| Goldman Sachs | Claude Code and GitHub limited to engineers; strats exempt |
| JPMorganChase | Outside AI agents would not be permitted inside the bank |
| Bank of America | Limits on agents’ access to consumer deposits while liability protocols are developed |
| Citigroup | Does not let third-party agents log into accounts or transact |
Jamie Dimon warned of “rogue agents” and said on a third-quarter earnings call last month that outside AI agents would not be permitted inside the bank. A day later, Bank of America placed limits on agents’ access to consumer deposits while it develops protocols for which party is liable for a payment an AI authorizes. PYMNTS reported that by Friday Citigroup did not let third-party agents log into accounts or transact. Who may let an outside system act is also the subject of JPMorgan and Amazon blocking external AI agents.
What Executives Should Take From the Reversal
This section is analysis, not an added Goldman disclosure. The restriction does not prove front-office coding failed. It shows a major bank ending a broad trial and keeping the tools where engineering is the job, with strats as the explicit exception.
Leaders weighing this against wide token budgets for legacy code rewrites can separate three decisions:
- Who may experiment on internal work.
- Who may use coding tools on client flows, positions, or external content.
- Who is accountable if output is wrong, exposed, or acted on.
Treat coding-tool access as a production right tied to role, not a culture perk. Strats staying in shows the cut can follow how a job is designed. Traders who already monitored flows and produced content are a change-management problem as well as a risk one: without a sanctioned path, the use cases the review studied do not disappear.
