Goldman Sachs has limited access to Anthropic’s Claude Code and Microsoft’s GitHub to developers, according to a policy that took effect in October. The tools had previously been available to a wider group of bank employees, including some traders.
The change has reportedly frustrated traders who used the tools to monitor customer trading flows and model market movements and positions, Bloomberg reported, citing confidential sources.
In July, Goldman CIO Marco Argenti had praised employees’ use of “vibe coding.” Staff had been using AI to build their own solutions before handing them to one of the bank’s 12,000 developers for refinement. A Goldman spokesperson said the bank now wants the tools used by those best placed to use them over the long term, citing more precise monitoring of productivity and risk management.
The restriction does not apply to “strats,” employees whose work spans trading, quantitative research and software development. Goldman’s in-house assistant, which combines models from OpenAI, Anthropic and Alphabet, remains available to all employees.
The move comes as banks reconsider AI spending and whether investments are generating measurable returns. According to benchmarking firm Evident, 12 of the 50 largest banks it tracks can currently attribute publicly forecast or achieved returns to their AI use. JPMorgan CFO Jeremy Barnum said in July that the bank planned to shift straightforward tasks, such as summarising analyst reports, to cheaper models, anticipating significantly higher token costs in the second half of the year.
Goldman is also involved in financing AI companies: it is lead underwriter for Anthropic’s planned IPO and is supporting Nvidia in large-scale credit transactions.
Sources: t3n.