Goldman halts Anthropic's Claude for Hong Kong bankers after weeks-long AI access blackout

Goldman Sachs has instructed its Hong Kong‑based bankers to cease using Anthropic’s Claude language model after a recent, unexplained interruption left employees unable to reach the firm’s broader suite of artificial‑intelligence tools for several weeks, a move announced in an internal memo circulated late last week that arrives at a moment when the bank’s Asian trading desks were already grappling with the practical consequences of an access blackout that, according to unnamed staff, rendered both client‑facing analytics and internal research applications effectively inert.

Company officials have offered no detailed justification beyond vague compliance concerns, a response that implicitly acknowledges a longstanding institutional gap between the rapid adoption of third‑party generative AI services and the rather slower, risk‑averse frameworks that govern their deployment across jurisdictions with divergent regulatory expectations; that the same compliance apparatus that permitted widespread experimentation with Claude elsewhere in the firm could, when pressed by an undefined Hong Kong regulator, retreat to a blanket ban highlights a procedural inconsistency that leaves senior bankers to rely on ad‑hoc workarounds while senior management continues to tout the bank’s commitment to cutting‑edge technology.

The episode, which underscores how a global financial institution can simultaneously champion artificial‑intelligence‑driven efficiency and yet be unable to assure uninterrupted access to the very tools it promotes, serves as a predictable illustration of the broader tension between innovation enthusiasm and the lingering, often opaque, governance structures that routinely curb such enthusiasm when regulatory uncertainty or internal risk assessments intervene, leaving observers to wonder whether the short‑term disruption to Hong Kong‑based teams will translate into a longer‑term reassessment of external AI partnerships, or merely become another footnote in the bank’s ongoing narrative of ambitious digital initiatives hampered by the perennial reluctance to reconcile speed with prudence.

Published: April 29, 2026

Published: April 29, 2026