1. News
  2. AI models & agents
  3. Goldman · AI tools
Goldman · AI tools2 min read

Goldman Sachs limits AI coding tools to its engineers

Goldman Sachs has limited AI coding tools to its engineers, a sign that software built by non-engineers near client data and money is getting a harder look.

By the Dashline deskSource: Hacker News (100+ points)
Goldman Sachs limits AI coding tools to its engineers
Photo: Hacker News (100+ points)

Goldman Sachs has stopped most of its traders and bankers from using AI coding tools at work. Guidance updated in October limits Anthropic's Claude Code and Microsoft's GitHub to the bank's engineers, Bloomberg News reported. The change followed a review of productivity and risk concerns.

What changed

Claude Code is an AI agent. It reads files, writes programs and runs them, typically with the same access as the person using it. Traders and bankers had used such tools to monitor client trading flows, model market moves and positions, and prepare written and graphic material. Some traders are frustrated, according to a person familiar with the matter.

Engineers keep access. Goldman has about 12,000 of them among roughly 47,000 staff. Strats, who work across trading, research and software, also keep it. The bank's in-house AI assistant, which uses models from several companies, is still available firmwide.

Why the bank may have done it

Code written by engineers usually goes through review, testing and deployment checks. A script a trader builds to watch client flows can end up in daily use with none of that. Banks must also keep information barriers between groups, such as dealmakers and traders. A tool built by one person with an agent was not designed around those controls.

Cost may be a factor too. Agent use is billed by the amount of text processed, and several companies that gave agents to everyone have seen spending climb. Bloomberg's sources said productivity was part of the review. Goldman itself said only that it makes certain AI tools available to staff to determine the best use cases.

The awkward timing

In July, Goldman's chief information officer, Marco Argenti, was praising "vibe coding" by non-technical staff. That means building software by describing what you want to an AI. He said traders and bankers had built tools and handed them to engineers to improve.

Goldman and Morgan Stanley are also lead underwriters on Anthropic's IPO, according to a filing dated October 26. Anthropic is expected to list on Nasdaq as ANT before the end of November. Most Goldman bankers marketing the shares can no longer use its coding agent.

What to ask in your own business

  • Who reviews software that staff build with AI?
  • Does any of it touch client data or money?
  • Who is paying for usage, and is anyone checking it?

Goldman has not said whether the limits are permanent or whether access could widen for trained staff.

The takeaway

If staff build their own tools with AI, make sure someone qualified checks that code before it touches client data or money.

More articles

You should really read these.

Next step

Let’s talk