Goldman Sachs AI plans face new scrutiny as banks tighten access rules

JPMorgan Chase has stopped Hong Kong staff from accessing Anthropic AI models, a move that lands squarely in the same lane Goldman Sachs employees are already watching: who gets access, where, and under what controls. For bankers, that is no longer a back-office technology question. It can change how teams draft client materials, summarize deal documents, and share work across borders.

Goldman already moved into this territory with a firmwide generative AI assistant launched in June 2025. Around 10,000 employees were already using the GS AI Assistant when the rollout was announced, and Goldman said the tool would help with summarizing complex documents, drafting initial content, and data analysis. In January, Marco Argenti said Goldman’s assistant would give employees access to approved models such as OpenAI’s ChatGPT, Google’s Gemini and Meta’s Llama, depending on the task.

That setup makes the governance questions more concrete for analysts, associates and coverage teams. Which jurisdictions are permitted, which use cases are approved, and what can be automated versus merely assisted all matter when a banker is pulling together a pitch book or preparing for a client call. The audit trail matters too, along with the basic question of where sensitive information goes once it enters a model. In a bank, productivity gains can disappear fast if compliance decides a workflow is too risky for a particular market.

Goldman has already shown that geography can override convenience. In April, the bank barred Hong Kong bankers from using Anthropic AI, and Anthropic said its Claude models had never been officially supported in Hong Kong. JPMorgan’s decision reinforces that this is becoming a jurisdiction-by-jurisdiction policy rather than a blanket firmwide default. For cross-border businesses like investment banking, private wealth and global markets, that means access to AI tools may be uneven by region and function, even inside the same franchise.

Goldman is also tied more directly to the vendor side of the market than many of its rivals. On May 4, Goldman Sachs, Anthropic, Blackstone and Hellman & Friedman announced a new AI-native enterprise services firm aimed at bringing Claude into core business operations. The venture was backed by Apollo Global Management, General Atlantic, Leonard Green, GIC and Sequoia Capital. That puts Goldman in an unusual position: using AI internally while also backing the ecosystem that is trying to sell it to enterprises.

For Goldman employees, the next AI rollout question is not whether the tools are powerful. It is whether the firm can keep access aligned with client confidentiality, regional rules and control requirements without slowing the front office to a crawl.