Goldman limits Claude Code and GitHub coding tools to engineers

Goldman Sachs has restricted Anthropic's Claude Code and Microsoft's GitHub coding tools to engineers, Bloomberg News reported via Bloomberg Law. Internal guidance updated in October, and seen by Bloomberg, followed a review of productivity and risk. The bank had allowed bankers to use the tools for months. Some traders were frustrated, Bloomberg reported; they had used the tools to speed work such as monitoring client trading flows and modelling markets. On Hacker News, commenters split. One camp said coding agents on desks with client data, email and the internet hit a known security risk, often called the lethal trifecta, and that banks are right to shut that down. Another focused on cost, saying enterprise Claude billing at API rates leads to overruns and that flat-rate rivals look attractive. Others speculated about an undisclosed incident or about IPO timing, since Goldman is linked to Anthropic's listing; those motives are commenter guesses, not confirmed in the Bloomberg report. A further camp said non-technical users dumping agent-built tools on engineering creates support and compliance risk, while some argued bankers should keep programmable tools if they run in safer sandboxes. No consensus. Sorted take: role limits on agents that write and run code are a sensible default in regulated firms, but collecting what traders used the tools for helps engineering rebuild the safe bits.