Goldman Sachs Restricts AI Coding Tools to Engineers

Estimated reading time: 6 minutes · Last updated: 2 November 2026

Goldman Sachs AI tools built for coding, including Anthropic's Claude Code and Microsoft's GitHub, are now reserved for the firm's engineers. For months before this, bankers and traders could use them too. The change is set out in internal guidance Goldman issued in October 2026, which became public on 2 November 2026. Goldman spokesperson Nick Carcaterra described it as the planned end of a trial in which the bank weighed productivity and risk management. The decision cuts against the bank's own July message about non-technical staff building software with AI. It also arrives as JPMorganChase, Bank of America and Citigroup set their own limits on AI agents.

After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term.

Nick Carcaterra, Goldman Sachs spokesperson, speaking to Bloomberg

What Goldman Changed, and Who Lost Access

Bloomberg News, which saw the internal guidance, reported on Monday, 2 November 2026, that Goldman Sachs has confined Claude Code, Anthropic's coding assistant, and Microsoft's GitHub to its engineering staff. Goldman published that guidance in October 2026. Until then, bankers had been free to use these tools in their work for several months.

The people losing access are not software developers. Bloomberg cited a person familiar with the matter who said some Goldman traders were frustrated by the change. Those traders had used AI to keep watch on client trading flows and to produce written material and graphics.

That detail shows what the tools had become on the trading floor. A coding assistant was doing analyst and presentation work, well outside the job it was designed for. The guidance pulls that informal use back inside the engineering function. Goldman has framed the change as a decision about who should hold the tools over time, rather than a verdict that the tools failed.

Why the Move Cuts Against Goldman's Vibe Coding Pitch

As recently as July 2026, Marco Argenti, Goldman's chief information officer, was promoting the opposite trend. In an interview he praised vibe coding by employees without technical backgrounds. Vibe coding, as Argenti used the term, meant non-technical staff building working software solutions with the help of AI tools.

Argenti described a two-step pipeline. Non-technical employees created solutions with AI, then handed them to one of the firm's 12,000 engineers to improve. The new guidance keeps the second step and removes most of the first. Engineers keep the tools, while the bankers who were prototyping ideas no longer have them.

Nick Carcaterra, a Goldman spokesperson, presented the change as the scheduled close of an experiment. "We make certain AI tools available to our people to determine the best use cases across different parts of the firm," he told Bloomberg. He added that the bank weighs factors such as productivity and risk management during that trial before choosing who keeps a tool for the long run. The explanation may well be accurate, but it sits awkwardly beside Argenti's public enthusiasm a few months earlier.

How JPMorganChase, Bank of America and Citigroup Drew Their Own Lines

Goldman is not alone in tightening rules, though the other big banks aimed at a different target. Jamie Dimon, chief executive of JPMorganChase, warned about "rogue agents" and, on the bank's third-quarter earnings call in October 2026, said JPMorganChase would bar AI agents built by third parties from operating within its walls. An AI agent is software that takes actions by itself, such as logging in or moving money, rather than simply answering questions.

Bank of America responded a day after Dimon's call with a condition instead of a ban. It is capping how far agents can reach into consumer deposits while it builds protocols to settle who carries liability when an AI approves a payment.

Citigroup followed before that week ended. The bank does not allow third-party agents to log into accounts or carry out transactions.

The difference matters. JPMorganChase, Bank of America and Citigroup are guarding the perimeter against agents brought in by customers or outside parties. Goldman's step is internal: it is choosing which of its own employees may use tools the firm itself supplies.

What the Restriction Says About AI Spending and Staffing

Behind all four decisions sits the same tension. Banks are spending millions of dollars on AI tools while trying to manage rising concerns about cost, productivity and cybersecurity risk. Every additional employee using outside AI software is another person whose work touches that software, and a bank weighing cybersecurity risk could reasonably want that group kept small.

Data from smaller companies points toward wider use, not narrower. Research from PYMNTS Intelligence found that 60% of small and medium-sized businesses using AI said it let them produce more without adding staff. PYMNTS argued that the bigger labour effect of AI may come from roles businesses never create rather than jobs they cut, and that demand for human judgment and accountability could grow as routine work gets cheaper.

Goldman's guidance suggests a large, regulated bank reads the trade-off differently from a small firm. For a small business, giving everyone AI is the whole point. For a bank that employs 12,000 engineers, concentrating coding tools in that group appears to be the preferred answer for now.

Arguments for and against Goldman's approach

The bottom line

Goldman's guidance turns a firm-wide experiment into an engineering perk. The bank says this is how its trials are meant to end, and Carcaterra's statement supports that reading. Yet the shift is real for traders who had folded Claude Code and GitHub into monitoring and presentation work, and it narrows the vibe coding model Argenti was praising in July 2026. Read alongside the October moves by JPMorganChase, Bank of America and Citigroup, a pattern emerges: large banks are keeping AI spending high while tightening who and what can act inside their systems. The unresolved questions are what Goldman's trial actually measured, and whether its rivals will apply similar limits to their own staff.

What to watch

• Watch whether Goldman restores AI tool access for bankers and traders in any form, or names replacement tools for them; no date has been set.
• Watch for Bank of America to finish its protocols on liability for payments an AI authorises; the bank has given no completion date.
• Watch whether JPMorganChase, Bank of America or Citigroup extend their limits from outside agents to their own employees' tools, as Goldman has; no date has been announced.