TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Goldman Sachs’s October internal guidance limits Anthropic’s Claude Code and Microsoft’s GitHub to engineers, restricting most bankers and traders from using the coding tools at work. The change follows a review that included productivity and risk concerns; the bank’s own AI assistant remains available firmwide.
Goldman Sachs has restricted Anthropic’s Claude Code and Microsoft’s GitHub to engineers, ending most bankers’ and traders’ workplace access to the coding tools, according to Bloomberg News. The October guidance followed a review of productivity and risk, and affects staff who had used AI tools to make models and monitor client trading flows.
The restriction applies to most bankers and traders, while strats—employees whose roles span trading, quantitative research and software engineering—still have access, according to a person familiar with the matter. The change covers third-party tools that can write and run code; Goldman’s in-house AI assistant, which uses models from multiple companies, remains available across the firm.
Some traders were frustrated by the change, a person familiar with the matter told Bloomberg. They had used the tools to track client flows, model market moves and positions, and prepare 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 guidance follows a period when Goldman had publicly discussed employees outside engineering using AI to build tools. In July, chief information officer Marco Argenti told Bloomberg that traders and bankers had made early versions and passed them to engineers for improvement. The bank has about 12,000 engineers, according to the source report. The new policy limits access to coding agents to that group, with strats retaining access.
Why Coding Access Is Narrowing
The change matters because code-generating agents can do more than suggest text: Claude Code can read files, write programs and run them, generally with the permissions of the person using it. On a trading desk, a quickly assembled script might become part of everyday work without the formal review, testing and deployment checks applied to engineering software.
That creates potential concerns for a regulated bank, including whether a tool handles unusual market conditions correctly and whether it respects client information controls and internal information barriers. Those barriers are designed to prevent sensitive information from moving between groups such as dealmaking bankers and traders. The source report describes these as reasons the policy may matter, not as confirmed incidents that prompted the restriction.
The policy also changes who can create early versions of work tools. Giving staff closest to a problem access to coding agents can speed experimentation, but limiting that access to engineers places more of the work inside established technical processes. Goldman’s review included productivity, Bloomberg reported; the bank has not publicly detailed how it weighed productivity against risk or whether costs influenced its decision.
Products Worth Considering
Apple AirPods Pro (1st Generation) with MagSafe Charging Case
Jackery Solar Generator 1000 V2 with 100W Solar Panel, 1024Wh LiFePO4 Battery
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz
As an affiliate, we earn on qualifying purchases.
From Employee Experiments to Controls
Goldman’s new rule is different from an earlier restriction in Hong Kong. In April, the bank removed Claude from its internal AI platform for bankers there. The Financial Times reported that Goldman had interpreted its contract with Anthropic strictly; an Anthropic spokesperson said Claude had not been officially supported in Hong Kong. That limitation was tied to location and contract terms, while the October rule is based on employee role. ChatGPT and Gemini remained available to Hong Kong staff at the time of the earlier restriction.
Goldman has not ended its work with Anthropic. Earlier this year, it said Anthropic engineers had spent six months at the bank helping develop Claude-based agents for trade accounting and client onboarding. Those back-office agents are run by Goldman engineering teams, according to the report. The bank’s relationship with Anthropic also has a separate financial dimension: Goldman and Morgan Stanley are named as lead underwriters on Anthropic’s IPO in the company’s registration filing, dated October 26.
Other financial firms have also set limits on agent access, though their policies differ. The source report says JPMorgan and Schwab are blocking outside agents from customer accounts. Barclays, meanwhile, said in October it expected half its developers to use Claude Code by year-end. These examples show firms drawing boundaries around where agents can operate, but they do not establish a single industry-wide approach.
““to determine the best use cases””
— Nick Carcaterra, Goldman Sachs spokesperson
Reasons and Scope Still Unclear
Goldman has not publicly provided a detailed explanation of the October restriction or said whether risk, productivity, cost or a combination drove the decision. Bloomberg reported that the bank reviewed productivity and risk concerns, but the source material does not specify the review’s findings or describe particular failures tied to staff use of the tools.
The precise boundaries of access are also not fully described. The report says strats retain access and that the tools are limited to engineers, but it does not list every eligible job title, explain how exceptions are granted, or say whether the rule covers all versions and uses of GitHub. It is also unclear whether Goldman plans to revisit the policy after further testing or review.
How Goldman Applies the Rule
The immediate next step is implementation of the October guidance across affected teams. Goldman has not announced a date for reviewing the restriction, a schedule for restoring access or a public process for employees to request exceptions.
The bank’s approach may become clearer through future statements about its internal AI tools and engineering controls. Separately, the source report said Anthropic was expected to list on Nasdaq under the ticker ANT before the end of November; that IPO timetable is distinct from Goldman’s workplace policy, and the supplied material does not confirm whether the listing occurred.
Key Questions
Which tools has Goldman Sachs restricted?
The October guidance restricts Anthropic’s Claude Code and Microsoft’s GitHub to engineers, according to Bloomberg News. The report characterizes the affected products as third-party coding tools that can write and run code.
Who can still use the coding tools?
Engineers retain access, and strats—staff working across trading, quantitative research and software engineering—also still have access, according to a person familiar with the matter. The report does not provide a complete list of eligible roles.
Can Goldman employees still use the bank’s AI assistant?
Yes. The restriction does not remove Goldman’s in-house AI assistant, which uses models from several companies and remains available across the firm, according to the source report.
Why did Goldman restrict access?
Bloomberg reported that the bank reviewed productivity and risk concerns. Goldman has not publicly explained the decision in detail, and the supplied reporting does not confirm a specific incident or identify cost as a reason.
Has Goldman ended its relationship with Anthropic?
No such change is reported. Goldman previously said Anthropic engineers helped develop agents for trade accounting and client onboarding, and the bank remains involved as a lead underwriter on Anthropic’s planned IPO, according to the source material.
Source: thestateofai.com
