Goldman Sachs updated internal guidance in October restricting access to advanced AI coding tools, specifically Anthropic’s Claude Code and Microsoft’s GitHub tools, to engineering staff, according to Bloomberg Law reporting.
Most bankers and traders reportedly lost access to those tools. The tools were reportedly limited to Goldman’s approximately 12,000 engineers.
One Group Gets a Carve-Out
The policy appears to exempt a distinct category of hybrid employees who sit at the intersection of trading and technical work.
The policy reportedly exempts “strats,” Goldman’s hybrid employees who blend trading, quantitative research, and software engineering in a single role.
That exception is worth noting. It suggests access decisions may track closer to technical accountability than to job title, though the available reporting does not confirm that rationale explicitly.
What Goldman Says It Was Doing
In Goldman’s own framing, the change reflects standard allocation after evaluation, not a withdrawal from AI.
Goldman spokesperson Nick Carcaterra described the change as a routine step rather than a reversal. According to FA Magazine, Carcaterra said the firm makes certain AI tools available experimentally, then puts them in the hands of those best suited to use them over the long term.
That sequence, broad access first and role-based allocation after evaluation, is how Goldman has characterized its approach publicly.
What Traders Were Actually Doing
Nontechnical employees had found practical, market-facing uses for the coding agents before the restriction took effect.
Access to coding agents was not merely a curiosity for nontechnical staff. Traders used the tools to monitor client trading flows, model market changes and positions, and generate written and graphic content, according to FA Magazine reporting based on an unnamed source.
The underlying workflow is what some in the industry call “vibe coding“: a trader could describe an idea in plain language, receive a rough prototype from the AI, and hand that prototype to engineers for production development. The appeal was direct access to fast prototyping, without a full engineering build cycle in between.
The Governance Problem With That Workflow
Coding agents that generate, modify, and test software raise accountability questions that carry particular weight in a regulated environment.
Coding agents create real governance questions around code review, data handling, and regulatory accountability. Security controls grow harder to standardize when the person writing the prompt is not the person accountable for the output in a regulated environment.
What the Reporting Does Not Tell Us
The specific findings from Goldman’s evaluation period remain undisclosed, leaving the precise rationale for the change unclear.
The available reporting does not establish what Goldman actually found during its evaluation period. Whether nontechnical users were less productive or created identifiable risk is unclear. So is whether the tools proved difficult to monitor or audit.
The restriction itself is strongly supported by multiple reports. The specific cause is not.
Where This Leaves the Industry
Goldman’s decision reflects a tension that financial firms across the industry are navigating as AI tools move from experimentation into governed deployment.
The restriction appears to place responsibility for these coding tools primarily with technical staff, with bankers and traders continuing to identify problems and shape requirements. Engineers become the principal users for translating those ideas into software using advanced AI agents. Importantly, one report indicates Goldman’s in-house AI tools remained broadly available across the firm, meaning this is not a general retreat from AI.
Broad access accelerates prototyping. Narrower access makes security controls and accountability easier to standardize. Goldman, at least for now, has chosen the second path for these particular tools.
