Goldman Walls Off AI for Its Own Dealers
As of 2026-11-02 18:14 UTC, On November 2, 2026, Bloomberg reported that Goldman Sachs has restricted its traders and bankers from using some AI tools. That single line is the most honest thing anyone in finance has said about generative AI this year.
The banks that spent the past two years telling investors AI would reshape the business just drew a line around the people who actually move the money. Traders and bankers are exactly the roles where a confident, plausible-sounding wrong answer costs real capital and invites real regulatory scrutiny. Goldman's restriction is not technophobia. It is a risk decision made by people who understand what an unlogged model output looks like in a compliance file.
The tension is structural. AI vendors sell horizontal capability; a trading desk needs a defensible audit trail for every input and output. Those are not the same product. A tool that helps draft a memo and a tool that sits next to position sizing cannot share the same governance tier, and Goldman is effectively admitting the industry has been treating them as if they could.
The part worth watching is what this does to adoption narratives. If the most sophisticated risk managers in banking are fencing off the front office, the productivity case for AI in finance narrows to back-office drafting and research summarization, not decision support. That is a smaller market than the pitch decks assume.
Goldman has not published the scope of the restriction, so the details of which tools and which desks remain unknown. But the direction is clear: in finance, the burden of proof sits with the model, not the trader.
