CVCOVANewsComputing, Opportunities, Value & Analysis
Cybersecurity & Privacy3 min read

Goldman Sachs restricts AI coding tools to its engineers: lessons for Latin American banking

According to Bloomberg, the bank took tools such as Claude Code, which they had been using for months, away from most of its traders and bankers. The case shows how a major financial institution is reorganizing the use of AI that writes code.

By COVA News newsroom
person using black and gray laptop computer

Photo: Kanchanara on Unsplash

For months at Goldman Sachs, a banker with no technical training could ask an artificial intelligence to write a program and see it running within minutes. For most employees, that is over. According to Bloomberg, the bank updated its internal guidelines in October and restricted AI coding tools to its engineers, including Anthropic's Claude Code and Microsoft's GitHub tools. Some outlets that picked up the story specify that this refers to GitHub Copilot, a detail Bloomberg does not specify.

The story, published around November 2, climbed to the top of Hacker News, the go-to forum for programmers, with 786 points and 416 comments. There is a reason for the interest: these are not simple chat windows. They are "coding agents," AI programs that do not just suggest instructions but write and execute them inside the company's systems.

Goldman spokesperson Nick Carcaterra attributed the move to "a variety of factors, including productivity and risk management," and said the bank offers certain tools to find the best use cases. According to The Decoder, he added that the experimentation phase is over. Goldman did not explain what specific risk prompted the change, and no source links it to regulatory requirements.

From celebrating vibe coding to setting limits

The shift is striking for how quickly it happened. In July, chief technology officer Marco Argenti praised so-called vibe coding, the practice of creating software by describing what you want in everyday language, without knowing how to program. Prototypes built by non-technical employees were then handed over to the bank's engineers, who number about 12,000, to be refined. In August, partner Chris Churchman warned that AI could erode bankers' reasoning skills. And as early as April, according to Bloomberg and the Financial Times, Goldman had restricted Claude for its staff in Hong Kong; Anthropic said the service was never officially supported there.

The move does not remove AI from the bank. According to the reports, Goldman's internal assistant, which combines several models, remains available to all staff. "Strats," roles that blend trading, quantitative analysis and programming, would reportedly keep their access, although only The State of AI and Gadget Review make that claim, based on an anonymous source. According to Bloomberg, some traders were unhappy about losing tools they used to monitor client flows and prepare text and charts.

There is also a business context: the relationship between the two companies is close. According to The State of AI, Anthropic engineers spent six months at Goldman building agents for trade accounting and client onboarding, which are now run by the engineering teams. And Goldman, along with Morgan Stanley, is among the leading banks working on Anthropic's initial public offering.

What the region can learn

There is no indication that the decision directly affects Latin America. The only bank in the region with a public agreement with Anthropic is Peru's Interbank, which announced in August a deal to use Claude Enterprise, the business version of the assistant, not necessarily its coding agent.

Even so, the case offers a useful takeaway, which is our own interpretation and not a rule: when an AI can write and execute code close to sensitive data, Goldman chose to leave it in the hands of those who have review processes and controlled deployment, and to route to engineering whatever originates outside it. It remains to be seen what the full text of the guidelines says, who made the decision and what specific risk drove it. For banks and companies in the region currently testing these tools, it is worth closely following those answers and reviewing what their own regulators require.

Related news

Cybersecurity & Privacy

An official Microsoft package stole credentials on PyPI: what to check if you use AI agents

Two versions of Microsoft's agent-framework reached PyPI carrying a credential stealer, one day after Google warned of attackers using AI to exploit flaws without human intervention. Here is what development and IT teams in Latin America should check right now.

Cybersecurity & Privacy

Microsoft and FAA outages, FBI and USDA breaches: no evidence of a coordinated attack

Over nine days, service outages and alleged intrusions coincided in the United States, and social media buzzed with talk of a “cyber pandemic.” The evidence points to separate technical failures. The proven danger for users is something else: fake emails that use the outages as bait.

Cybersecurity & Privacy

Nginx: two critical HTTP/2 flaws expose millions of servers; here's how to check and protect yours

F5 has fixed two vulnerabilities in Nginx that allow an attacker to take control of a server without a username or password. Public proof-of-concept code is already circulating. This guide explains how to tell whether your server is at risk, which version to update to and what to do while you wait for your system's patch.