Goldman Sachs Pulls Back AI Coding Tools From Traders and Bankers After Risk Review

Goldman Sachs Restricts AI Coding Tools Access

Goldman Sachs has barred most traders and bankers from using AI coding tools, including Anthropic's Claude Code and Microsoft's GitHub, according to internal guidance updated in October1

1

. The restriction limits these tools to engineers after the bank conducted a comprehensive review of productivity and risk concerns2

2

. Goldman Sachs spokesperson Nick Carcaterra explained that after a period of experimentation assessing factors including productivity and risk management, the bank decided to place the tools "in the hands of those who are best suited to leverage them for the long term"1

1

.

The directive does not cover "strats," employees whose roles combine software engineering, quantitative research and trading1

1

. This decision marks a significant retreat from the Goldman Sachs AI policy that Chief Information Officer Marco Argenti promoted just months earlier in July, when he spoke favorably of "vibe coding" among non-engineers1

1

.

Frustration Among Traders and Bankers

The restriction has frustrated some traders who had turned to AI tools for traders and bankers to perform critical tasks1

1

. These professionals used the tools for monitoring how clients were trading, simulating shifts in markets and positions, and drafting text and visuals1

1

. Traders relied on these AI coding tools to watch client flows, model positions and produce written and graphic material5

5

.

The policy shift undermines comments from Marco Argenti, who praised nontechnical Goldman staff using AI to build software prototypes before passing them to engineers for refinement2

2

. Under the previous approach, bankers and traders could create their own solutions using AI, then hand the results to an engineer from the bank's pool of 12,000 for polishing1

1

.

Production Readiness and Risk Management Challenges

The core issue driving restricting access to AI tools centers on production readiness rather than the technology itself. AI has made software remarkably easy to start building, but what non-engineers cannot produce by the end of the afternoon is something IT will put into production3

3

. Anything touching client flows at a regulated institution must be secure, resilient and auditable, integrated with core systems and signed off by compliance3

3

.

The handoff becomes the bottleneck in AI adoption in finance. Business users produce prototypes, and engineers inherit the job of making them production-ready, causing backlogs to grow faster than any team can clear3

3

. Financial software cannot be judged only by whether it runs—an application can produce plausible numbers while applying wrong pricing assumptions5

5

. A coding agent with access to internal systems can also pull confidential data, link restricted databases or push changes outside the bank's approval process5

5

.

Concerns About Skill Degradation

Goldman partner Chris Churchman warned in August that the spread of AI across Wall Street risks degrading the analytical abilities of the next generation of bankers1

1

. Churchman, who co-chairs the bank's Global Banking and Markets AI working group, said Goldman had not yet determined how it would manage that transition regarding skill degradation1

1

. The technology could degrade bankers' reasoning skills if relied on too heavily by employees4

4

.

How Other Financial Institutions Handle AI Tools

Other banks are working out different approaches to how employees should use AI. JPMorgan Chase said in July it wanted staff to match each task to a suitable model, saving the priciest option for harder work rather than something like condensing an analyst report1

1

. JPMorgan Chase Chief Information Officer Lori Beer noted that AI is evolving so quickly that it can be difficult to assess risks in real time, stating "we want to move at speed, but sometimes we must pull back to get controls in place"2

2

.

BNY Mellon takes a different approach, letting employees outside engineering build on its Eliza platform inside an environment designed around the bank's data, legal and compliance requirements5

5

. BNY lists more than 400 AI-enabled solutions in production, and artificial intelligence now writes roughly 40% of its code5

5

.

Mitsubishi UFJ Financial Group built an in-house alternative to Anthropic's Claude Code after finding the tool could work around its information-access limits and compliance safeguards5

5

. Deutsche Bank keeps tools with professional developers, with more than 15,000 using Microsoft's GitHub Copilot and Google Gemini Code Assist, saving an estimated 1.5 to 2.5 hours weekly5

5

.

Future of AI Adoption in Banking

Among executives at firms already using or piloting AI agents, 43% would give agents moderate system access with human approval, and none would grant full access, according to a July 2025 survey of 60 CFOs at U.S. companies with at least $1 billion in revenue5

5

. Among financial services and insurance firms, 85% plan to increase AI budgets over the next 12 months, with 30% naming data quality as their biggest barrier5

5

.

Goldman Sachs has created its own AI assistant drawing on models from OpenAI, Anthropic and Alphabet1

1

. The bank has limited AI tools before—earlier this year, Goldman restricted staff in Hong Kong from using Anthropic's Claude amid growing tensions between the US and China over AI1

1

4

4

. The dividing line on Wall Street is not whether AI should write code, but who can authorize what that code does, which systems it can reach and who is accountable when it goes wrong5

5

.