Goldman Sachs narrowed access to Anthropic’s Claude Code and Microsoft’s GitHub AI coding tools in guidance updated in October, after allowing bankers to use them for months. The bank said its review considered productivity and risk management; the reported change restricts these tools to engineers, rather than establishing a ban on all AI use.
1
5
6
Why Goldman narrowed access
Goldman spokesperson Nick Carcaterra described the decision as following a period of experimentation and an assessment of factors including productivity and risk management. He also said the tools would be put in the hands of people best suited to use them over the long term. The public explanation does not specify which particular risks or productivity findings drove the change, or provide measurements of either.
26
32
That makes the decision best understood as a change in who can use particular coding tools—not proof that Goldman found AI coding ineffective or ended its wider AI work. The available reporting does not describe a firm-wide prohibition on AI.
1
6
What the restriction changed for bankers and traders
Before the guidance changed, bankers and other employees had used the tools for work. Reporting says some traders used them to monitor client trading flows, model market changes and positions, and create written and graphic material. The restriction cut off access for most bankers and traders, frustrating some staff, according to reports citing Bloomberg.
5
7
27
The change is significant because these tools had been used beyond software engineering. But the evidence provided does not establish whether employees can still complete the same tasks through other approved tools or processes.
Who retained access—and how that differs from earlier experimentation
Reports describe an exception for “strats,” a group whose work combines quantitative or trading responsibilities with software engineering. That description points to a role-based distinction: access was retained for people whose jobs include engineering work, rather than for everyone who might find coding useful. The reporting does not spell out every eligibility rule.
18
20
The restriction also contrasts with Goldman CIO Marco Argenti’s earlier public support for nontechnical employees using AI to make prototypes that engineers could develop further. That earlier encouragement of experimentation helps explain why the narrower access policy may feel like a reversal, but it does not establish that Goldman abandoned prototyping or AI adoption altogether.
4
29
What the reporting does—and doesn’t—show
The clearest stated rationale is a review involving productivity and risk management. The sources provided do not establish cybersecurity, cost, or compliance concerns as specific causes of this decision, nor do they verify broader comparisons with other banks’ geographic restrictions.
1
26
The takeaway is narrower than a general “AI ban”: Goldman moved advanced coding tools into a more limited, engineering-focused pool after a trial period. The public explanation identifies the categories considered, but leaves the detailed findings and the boundary of access unclear.