Goldman Sachs restricts Claude Code and GitHub to its engineers

Goldman Sachs has restricted Anthropic's Claude Code and Microsoft's GitHub to its engineers, according to a report by Bloomberg News relayed by State of AI. Internal guidance updated in October limits the tools to the bank's engineers after a review of productivity and risk concerns. As a result, most traders and bankers at the bank can no longer use AI coding tools at work. Strats, who work across trading, quantitative research and software engineering, still have access, according to a second person familiar with the matter.

The change has frustrated some traders, a person familiar with the matter told Bloomberg. They had used the tools to monitor client trading flows, model market moves and positions, and put together written and graphic material. Goldman spokesperson Nick Carcaterra said the bank makes certain AI tools available to staff "to determine the best use cases" across the firm.

The reversal stands out because, as recently as July, chief information officer Marco Argenti was talking up "vibe coding" by Goldman's non-technical employees. He told Bloomberg that traders and bankers had built tools with AI and passed them to the firm's engineers, of whom there are about 12,000, to be improved. At a conference in Turin last month, Argenti described the bank's AI adoption in waves and said developers, more than 12,000 of the bank's roughly 47,000 staff, had made up the first. The October guidance limits the coding agents to that group.

There is an earlier restriction with a different basis. In April, Goldman removed Claude from its internal AI platform for bankers in Hong Kong. The Financial Times reported that the bank had read its contract with Anthropic strictly, and an Anthropic spokesperson said Claude had never been officially supported in Hong Kong. ChatGPT and Gemini remained available there. JPMorgan removed Claude from its Hong Kong approved list in June. The Hong Kong restriction was based on location and contract terms; the new one is based on role. Goldman's in-house AI assistant, which uses models from several companies, is still available across the firm. The restriction applies to third-party agents that can write and run code by themselves.

Goldman is still working closely with Anthropic. Earlier this year it said Anthropic engineers had spent six months at the bank helping build Claude-based agents for trade accounting and client onboarding; those agents sit in the back office and are run by Goldman's engineering teams. Goldman and Morgan Stanley are also the lead underwriters on Anthropic's IPO, according to the S-1 filed October 26. Anthropic is expected to list on Nasdaq under the ticker ANT before the end of November. In State of AI's words, over the next few weeks Goldman's bankers will be marketing shares in a company whose coding agent most of them are no longer allowed to use.

State of AI offers its own reading of why a coding agent is riskier on a trading desk. Claude Code reads files, writes programs and runs them, typically with the same access as the person using it. Engineers' code usually passes through review, testing and deployment controls; a script a trader builds to watch client flows can reach daily use without any of them. Banks must also keep information barriers between groups, for example between bankers working on deals and traders who could profit from knowing about them, and Goldman built its internal AI platform around such controls. A tool a trader assembles with an agent was not designed with them in mind. There is also the question of whether anyone understands the code: the article cites a Meta engineer who said last month that he now ships code he can't fully account for. Cost may also have played a part, the article suggests, since agent usage is billed by the token; it points to Amazon shutting down an internal leaderboard after employees ran agents on pointless tasks to climb it, and to Accenture now funding $29 a month of Claude per employee. Bloomberg's sources said only that productivity was part of Goldman's review.

The article contrasts Goldman with Barclays, which said in October that it expects half its developers to be using Claude Code by the end of the year, and three weeks later named developers among 5,000 job cuts. Like Goldman, Barclays is giving coding agents to engineers; unlike Goldman, it is also using them to reduce how many engineers it employs. Elsewhere, JPMorgan and Schwab are blocking outside agents from customer accounts.

In its "Our take" section, State of AI argues that Goldman was right to restrict the tools and wrong in how it went about it. Requiring that agent-written code come from people trained to review it is a sensible rule for a regulated business, it says, but withdrawing access through a guideline update, with no replacement mentioned in Bloomberg's reporting, tells the traders who found good uses that they were the risk.

Key facts

Internal Goldman Sachs guidance updated in October restricts Claude Code and GitHub to the bank's engineers, per Bloomberg News, after a review of productivity and risk concerns.
Most traders and bankers lose access; strats, who work across trading, quantitative research and software engineering, keep it. The in-house AI assistant remains available firm-wide.
In July, CIO Marco Argenti was praising AI-built tools from non-technical staff; the bank has about 12,000 engineers out of roughly 47,000 staff.
Separately, Goldman removed Claude from its Hong Kong bankers' AI platform in April, a restriction based on location and contract terms rather than role.
Goldman and Morgan Stanley are lead underwriters on Anthropic's IPO, expected on Nasdaq as ANT before the end of November, per the S-1 filed October 26.

Why it matters

A major bank has drawn a line between engineers and everyone else on who may run agents that write and execute code. It is a concrete case of a regulated firm tightening access after earlier enthusiasm: in July the CIO was promoting trader-built tools. The article places it against a mixed picture in banking. Barclays expects half its developers to use Claude Code by year end and has also named developers among 5,000 job cuts, while JPMorgan and Schwab are blocking outside agents from customer accounts. The timing is awkward too, since Goldman is a lead underwriter on Anthropic's IPO.

Who it affects

Most of Goldman's traders and bankers, some of whom were using the tools to monitor client flows, model market moves and positions, and prepare written and graphic material. Goldman's engineers (about 12,000, per the CIO) and strats keep access. Bankers working on the Anthropic IPO are affected in a symbolic sense: the article notes they will market shares in a company whose coding agent most of them can no longer use. Leaders at other companies deciding how widely to hand out agents are the article's wider audience.

How to use it

The practical advice comes from State of AI's own "Our take", not from Goldman. Settle who owns agent-built code before handing agents out: when someone outside engineering builds a tool with Claude Code or Copilot, a named engineer should take responsibility for it before it touches production data. Make existing controls cover agent-written code, since information barriers, data entitlements and software approval rules still apply when the code came from a prompt. And when a tool is taken away, record what people were using it for; Goldman's traders were watching client flows and modeling positions, and they will still need to do both.

How solid is it

The core claim is Bloomberg's, relayed secondhand by State of AI, which gives neither the guidance text nor a date within October. Details such as the frustration of some traders and strats keeping access rest on unnamed people familiar with the matter. Goldman's only on-record comment, from spokesperson Nick Carcaterra, is about making certain AI tools available "to determine the best use cases", and does not explain the October change. The Hong Kong episode in April was reported by the Financial Times, with an Anthropic spokesperson saying Claude was never officially supported there. The IPO details come from the S-1 filed October 26, and the listing is an expectation, not a confirmed date.

Risks and caveats

The reasons for the restriction in the article (unreviewed code, information barriers, code nobody fully understands, possible cost) are State of AI's analysis, not Goldman's stated reasons; Bloomberg's sources said only that productivity was part of the review. The article does not say how many traders and bankers lost access, and Bloomberg's reporting mentions no replacement tool for them. The article does not say Goldman is scaling back its work with Anthropic; the back-office agents built with Anthropic engineers continue. The Hong Kong restriction (location and contract terms) and the October one (role) are described as different in basis. The final verdict, that Goldman was right to restrict but wrong in method, is the publication's opinion.

“to determine the best use cases”

— Nick Carcaterra, Goldman spokesperson, on why the bank makes certain AI tools available to staff