Goldman Sachs restricts traders’ access to AI coding tools

From experimentation to restricted access

Goldman Sachs spokesperson Nick Carcaterra told Bloomberg the decision reflected “various factors, including performance and risk management.” He said the experimentation phase was over and access would be limited to employees “best suited for long-term work with them.”

In July, chief information officer Marco Argenti had spoken approvingly of employees without technical backgrounds using AI to turn ideas into prototypes. Now, traders and asset managers appear to have less room to try that approach themselves.

What the company has not said is which tools are affected, how many employees will retain access, or what performance and risk concerns drove the change. Those details matter: restricting access could mean the prototypes were not useful enough, or that Goldman wants a smaller group to manage the risks. The announcement does not distinguish between those possibilities.

I think the key shift is less about whether Goldman still uses AI coding tools than about who gets to experiment with them. The company is moving from broad access toward a narrower group, but without explaining how it will judge who is suited to keep using them. That leaves the original promise—ideas from business teams becoming prototypes before engineers take over—dependent on a gate the bank has not described.