Goldman Sachs Restricts Claude Code and GitHub to Its 12,000 Engineers, Cutting Off Traders and Bankers
- Goldman Sachs guidance updated in October limits Anthropic's Claude Code and Microsoft's GitHub to the bank's roughly 12,000 engineers, cutting off traders and bankers who had used the tools to monitor client trading flows, model positions and build written and graphic material.
- The restriction followed a review of productivity and risk: an agent like Claude Code reads files, writes programs and runs them with the same access as the user, so a trader's script can reach daily use without the review, testing and deployment controls engineering code goes through, and without the controls behind Goldman's internal AI platform's information barriers.
- In April Goldman removed Claude from its internal AI platform for Hong Kong bankers, with the FT reporting the bank read its Anthropic contract strictly; JPMorgan removed Claude from its Hong Kong approved list in June. The new limit is based on role rather than location.
- Goldman and Morgan Stanley are lead underwriters on Anthropic's IPO, per the S-1 filed October 26, with Anthropic expected to list on Nasdaq under the ticker ANT before the end of November.
- Goldman's in-house AI assistant and the back-office trade accounting and client onboarding agents that Anthropic engineers spent six months building at the bank remain in use, and strats across trading, quantitative research and software engineering keep access.
Hacker News 의견들
Original source is Bloomberg via fa-mag, and this State of AI site looks like an AI-written thing run by a spam company. Somehow it skates right past uBlock too, which is weird.
A trading floor is a bad place for unreviewed code written by people who can't read it. The information barrier problem alone would be reason enough, and requiring agent-written code to come from people trained to review it is a sensible rule for a regulated business. Jokes on them though, devs are losing the capacity to read the code too.
At which point it's vibe coding, not engineering.
Worse, the instruction now is to keep prompting and not even look at it, let the software factory review and regenerate when quality drops below 0.8.
I have a lot more sympathy for non-technical users writing slop than for engineers. It's great that more people get to use a flexible programmable environment instead of a fixed set of programs, and if security is the issue, the interesting problem for the engineering team is designing an environment where bankers cannot hurt themselves.
Every company is going to end up with an interesting ecosystem of internal APIs. Lots of startups promise this future but none are technically equipped to deliver it.
Designing an environment where bankers cannot hurt themselves sounds like the web platform.
In finance it's probably compliance more than punishing non-technical users. If a model trains on your internal financials and someone else uses it to make a quick buck, that's insider trading.
If security is the issue they need the bulletproof platform first, then give the less technical users access. And rebellion isn't the right framing, the bankers at an investment bank decide where the company goes anyway.
I wonder if the medium term shift is away from Claude Code, which runs on your own machine and is risky when you have client information, toward something sandboxed like managed agents. Or just the Claude app with local execution disabled by org policy.
A banker being able to ship an app other people run with no IT review is a compliance problem by itself. Imagine a banker builds a tool that lets him talk to a trader across the Chinese wall without touching monitored email, Slack, phone or Zoom.
Or an idiot writes an app that scans his email, summarizes client info he's privy to and serves it to the trading desk, not knowing entitlements exist. Summarizing email is a huge use case for non-technical users in finance.
Probably it's just overwhelming slop. PoCs and 1000-page Claude artifacts instead of a 10-line answer in chat, drowning everyone.
How does 'bankers cannot hurt themselves' not end up right back at take-it-or-leave-it programs? Pile on enough abstractions that must be adhered to and you get the same thing. And security is about data too: bankers upload sensitive data to AI every day.
Who maintains all these bespoke apps? Banks aren't tech giants, most engineering teams don't bring in money and headcount is strictly managed.
I've been on the receiving end of these trader-built tools. 'Improved' is doing heavy lifting, completely overhauled to be operationally viable, or more likely thrown into the dark recesses of the backlog.
Serious question, are you any better off than if the traders had just forwarded you the prompt directly?
I have a non-technical friend building apps for a much smaller investment bank, ingesting old pitch decks and deal documents. He's gotten seriously into CI and SDLC, but I don't think most management is on that page.