Goldman Sachs is reportedly restricting workers from using advanced artificial intelligence tools for regular work, including coding. The move follows an internal evaluation of productivity and risk management concerns.
This marks a notable shift for the company, which had earlier encouraged employees to experiment with AI. Notably, earlier this year, the financial firm partnered with Anthropic with an ambition to build AI agents that could automate regular tasks within the bank.Now, a Bloomberg report reveals that following an internal guidance update in October, Goldman is limiting access to AI coding tools such as Anthropic's Claude Code and Microsoft's GitHub to engineers.
Employees had been allowed to use these tools for months. Reacting to the development, Replit's CEO took to X (formerly Twitter) and said that this trend was widespread in finance, warning that a “ban hammer is coming down.”The decision has reportedly frustrated some traders who had been using these tools to speed up everyday tasks.
These included monitoring client trading activity, modelling market changes and positions, and creating written and graphic content.Interestingly, as recently as July, Goldman CIO Marco Argenti had spoken positively about the idea of “vibe coding” among non-technical employees, including traders and bankers. These employees were using AI tools to build solutions for their work before handing them over to the bank's team of around 12,000 engineers for further improvements.
A Goldman Sachs spokesperson told Bloomberg that the bank had allowed employees to experiment with different AI tools to identify useful applications. However, after assessing factors such as productivity and risk management, it decided to put these tools in the hands of employees best suited to use them in the long run.The latest restriction does not apply to so-called “strats”, employees who work at the intersection of trading, quantitative research and software engineering.Wall Street is still figuring out its AI strategy Goldman Sachs is not alone in trying to strike a balance between AI adoption and its potential risks.
Banks are spending millions of dollars on AI tools, while also dealing with concerns around cybersecurity, costs and productivity.JPMorgan, for instance, said in July that it wanted employees to understand which AI model was best suited for a particular task. This could help avoid spending money on expensive models for relatively simple work, such as summarising an analyst report.Goldman Sachs itself has developed an AI assistant that combines models from OpenAI, Anthropic and Google.- Ends