In July, Goldman Sachs's chief information officer was publicly praising non-engineers who built tools with AI. Four months later, Bloomberg reports that most of those same people have lost access to the tools. On Monday, November 2, 2026, Bloomberg said Goldman has limited advanced AI coding tools, including Anthropic's Claude Code and Microsoft's GitHub offerings, to its engineers.
It is a small policy change at one bank, but it lands in the middle of a bigger question every employer is facing: when staff can ask AI to write code and build things, who should be allowed to, and who is responsible when it goes wrong?
Quick summary: Per Bloomberg-based reports on November 2, 2026, Goldman Sachs restricted Claude Code and GitHub AI tools to engineers under internal guidelines from October. A spokesperson cited productivity and risk management and, per the report's paraphrase, said the experimentation phase is over. Goldman has not published the policy, and the number of affected staff is not disclosed.
What was reported
According to The Decoder's summary of the Bloomberg report, Goldman has cut most employees off from advanced AI coding tools; only engineers keep access. The restriction comes from internal guidelines dated October. Traders and bankers had used the tools for months to monitor trading flows and create content, and the rollback reportedly frustrates some of them. (Source: The Decoder)
PYMNTS adds that traders had used AI to monitor client trading flows and to create written and graphic materials. (Source: PYMNTS) We could not open Bloomberg's original article, so this relies on secondary summaries, and outlets describe the exact scope a little differently.
The July reversal
The timing is what makes the story notable. In July, CIO Marco Argenti praised "vibe coding" by non-technical staff. The intended workflow, according to The Decoder, was to prototype with AI and then hand the result to an engineer to scale into production, and that plan is no longer in place. PYMNTS puts the size of that engineering bench at 12,000 engineers. (Sources: The Decoder, PYMNTS)
What Goldman says
Spokesperson Nick Carcaterra cited a "variety of factors including productivity and risk management," per The Decoder, and said, in the report's paraphrase, that the experimentation phase is over and that access is being limited to people "who are best suited to leverage them for the long term." Bloomberg's report did not give further reasons, so anything beyond that is speculation.
| Bank | Reported stance | Reported by |
|---|---|---|
| Goldman Sachs | Claude Code and GitHub AI tools limited to engineers | Bloomberg, via The Decoder and PYMNTS |
| JPMorgan Chase | Outside AI agents will not be permitted inside the bank, per CEO Jamie Dimon, who warned of "rogue agents" | PYMNTS |
| Bank of America | Conditions rather than a ban; limits agents' access to consumer deposits while it develops liability protocols | PYMNTS |
| Citigroup | Does not let third-party agents log into accounts or transact | PYMNTS |
Note the distinction: the other three banks' stances concern outside AI agents acting on customer accounts, a different issue from staff using coding assistants internally. The common thread is caution as AI moves from chat windows to systems that act.
Why banks might tighten up
Reporting does not say which risks drove Goldman's decision, so here are general reasons any regulated firm might limit coding agents, labeled as our analysis rather than Goldman's stated rationale:
- Code written by non-engineers may skip the review, testing and security checks that engineering teams enforce.
- Regulated firms must be able to explain and audit the systems that touch trading, client data and reporting.
- Giving an agent broad file and system access widens what can go wrong if it makes a mistake.
- Standardizing on a few vetted tools is easier to govern than letting every desk experiment.
What it means for your workplace
If your employer has not set rules for AI coding tools, expect them soon. The Goldman case shows a pattern: broad experimentation, then a tightening around people with the training to own the output. Individuals can get ahead of it by learning what these tools do, keeping sensitive data out of public tools, and asking IT what is approved.
Before using any AI coding tool at work, check your employer's policy, never paste confidential client or trading data into a personal account, and have an engineer review anything that will touch production systems.
01What did Goldman Sachs change?
Per Bloomberg-based reports on November 2, 2026, it limited advanced AI coding tools such as Claude Code and Microsoft's GitHub tools to engineers.
02Why did it do it?
A spokesperson cited productivity and risk management and, per the report's paraphrase, said the experimentation phase is over. No further reasons were given in reports we reviewed.
03How many employees are affected?
The number has not been disclosed in the coverage we reviewed.
04Is Goldman abandoning AI?
There is no evidence of that in the reports; the change is about who can use coding tools.
05Does this mean Claude Code is unsafe?
No. Nothing in the reports says the tool failed; the decision was about who should use it and about risk management.
Deciding which AI tools fit your own work? LumiChats lets you try 40+ premium models, including Claude, GPT and Gemini, in a single subscription, so you can compare them on non-sensitive tasks before your team sets a policy.
