Goldman Sachs Group Inc. has abruptly restricted access to advanced artificial intelligence (AI) coding tools across most of its workforce, reversing a permissive policy that had allowed non-technical staff to build custom software for daily financial operations.
Under revised internal guidance issued in October, only designated software engineers and quantitative “strats” — specialists bridging trading and development — may now deploy Microsoft Corp.’s GitHub and Anthropic’s Claude Code. The policy removes the tools from the hands of investment bankers and traders, several of whom voiced frustration after relying on the software to model positions, track client flows, and draft content.
The restriction marks an abrupt departure from the bank’s posture just months earlier. In July, Goldman Chief Information Officer Marco Argenti praised grassroots vibe coding among front-office professionals, noting that non-engineers were independently assembling functional solutions before handing them off to the firm’s 12,000 professional developers for refinement.
Earlier in the year, Goldman President John Waldron underscored the firm’s broader automation ambitions, describing the enterprise as a “human assembly line” primed for technological disruption.
The sudden shift illustrates Wall Street’s delicate balancing act: capturing the productivity promises of generative AI while managing acute security, legal, and regulatory exposures. AI coding agents operate under the permissions of their users. On sensitive trading desks, automated code execution presents substantial governance hurdles, from safeguarding material non-public information and maintaining data barriers to auditing unvetted machine-generated code.
“We make certain AI tools available to our people to determine the best use cases across different parts of the firm,” said Goldman spokesperson Nick Carcaterra. “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.”
Similar caution is surfacing across the financial sector. Peers like JPMorgan Chase & Co. have tightened oversight, guiding staff on cost management to avoid running expensive frontier models for basic workplace automation.
Goldman itself continues to maintain proprietary platforms — including an internal assistant synthesizing models from OpenAI, Alphabet Inc., and Anthropic — while previously imposing regional guardrails, such as limiting Claude usage in Hong Kong.
The operational clampdown arrives alongside an ironic corporate juxtaposition: Goldman is functioning as the lead underwriter for Anthropic’s planned initial public offering, while coordinating debt deals for foundational AI infrastructure suppliers such as NVIDIA Corp.
For the traders and bankers left without access, the abrupt withdrawal has created workflow bottlenecks with no immediate internal replacement. Industry observers note that while centralizing coding access addresses critical data compliance, firms will increasingly need formal engineering channels to support front-office teams whose operational habits are now firmly tethered to autonomous software.
