Industry News
Goldman Sachs took back the Claude Code from its engineers, and the traders who relied on it to monitor market trends stopped using it immediately.
Goldman Sachs recently revised its internal guidelines. Previously, tools such as the Claude Code and related Microsoft GitHub resources (like GitHub Copilot) were also available to bankers; however, these tools are now primarily restricted to engineers.
Among those who were affected, there was a group of traders. Previously, they used these tools to monitor their clients’ trading activities, simulate market trends and changes in their positions, and also generate written reports and visual charts as a byproduct of their work. However, once their access to these tools was restricted, some of them became dissatisfied with the new arrangements.
There is one group of people who were not affected at all: those employees who work in the areas that overlap trading, quantitative research, and software engineering.
This line clearly illustrates the point. The basis for Goldman Sachs to assess risk does not lie in whether you use AI or whether you have the ability to review what it generates. If a trader uses an intelligent system to write scripts for position simulation, he may not be able to determine whether the resulting numbers are accurate. However, engineers and strategists can certainly do so.
Banks also face significant compliance pressures in this regard. Model risk management is subject to regulatory requirements; if an unreviewed script affects the decision-making process regarding actual transactions, it will be extremely difficult to determine who is responsible in the event of any problems occurring.
Spokesperson Nick Carcaterra’s response was quite formal: “We will evaluate the potential uses of these systems in various departments through testing, take into account both productivity and associated risks, and then decide which options are suitable for long-term use.” In other words, this round of restrictions may not represent the end of the process.
The company is not opposed to the use of AI at all. Goldman Sachs has already implemented its own AI assistants throughout the company, which can be used for processing documents, drafting content, and analyzing data. The problem lies in the fact that traders actually need intelligent systems that are capable of running simulations manually; it remains to be seen whether these internal assistants can meet such requirements. If this gap is not addressed, alternative solutions will inevitably emerge in the future.