Goldman Sachs Hong Kong employees restricted from using Claude, Anthropic admits: Never supported HK region, AI geopolitical restrictions spread

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1777 words

Financial Times exclusive, confirmed by Bloomberg on April 29: Goldman Sachs Hong Kong employees lost access to Claude several weeks ago. The move followed consultations between Goldman Sachs and Anthropic, resulting in a strict contractual interpretation confirming that Hong Kong-based employees are prohibited from using any Anthropic products. An Anthropic spokesperson also stated for the first time that the Claude model has "never been officially supported in Hong Kong." This is more than just an IT policy adjustment at a single bank—it reveals the service boundaries of AI providers amidst US-China geopolitical tensions, and the quiet contraction of toolchains within foreign financial institutions (including crypto divisions) in Hong Kong. (Context: Anthropic Mythos's power triggers emergency meeting: Bessent and Powell convene Citi, Goldman Sachs, Bank of America, JPMorgan, and Morgan Stanley; five major banks focus on financial risks.) (Background: Anthropic requires real-name KYC verification! Certain Claude features will require uploading identity documents, escalating compliance pressure.) Employees at the Hong Kong branch of Goldman Sachs quietly lost access to a tool—Anthropic's Claude—without any official notification. A Financial Times exclusive on April 28, confirmed by Bloomberg the following day, reported that Claude has been completely disabled for Hong Kong employees on Goldman Sachs' internal AI platform, with the block covering all use cases including coding assistance and document analysis. Other mainstream models—Google Gemini and OpenAI ChatGPT—continue to function normally on the platform; only Claude is absent. The incident was not a technical glitch, but a corporate compliance decision. After consulting with Anthropic, Goldman Sachs adopted a strict interpretation of their contract, concluding that Hong Kong employees fall outside the scope of Anthropic's service and are therefore prohibited from using any Anthropic products. Anthropic's official response was brief and clear—a spokesperson told the FT that the Claude model "has never been officially supported in Hong Kong" and declined further comment. There is a substantive basis for this statement: the list of supported regions officially published by Anthropic has never included Hong Kong, and Claude has long been blocked at the IP level in the region. This restriction is not required by Hong Kong regulations, but is a proactive policy choice by Anthropic. The first layer is the legal environment. Following the implementation of the National Security Law in 2020, the legal framework in Hong Kong has created uncertainty regarding data sovereignty and compliance obligations for US tech companies, leading Anthropic to adopt a conservative approach. The second layer is US government relations. Anthropic holds numerous US federal government contracts covering defense and intelligence sectors, and its geopolitical control policies must align with the US export control framework. Anthropic itself is a vocal advocate for US advanced chip export controls. The third layer is model security concerns. In early 2026, Anthropic publicly accused Chinese AI labs, including DeepSeek, of systematically extracting Claude's capabilities through over 16 million interactions via "model distillation." This incident further reinforced Anthropic's stance on access restrictions in China-related regions. Notably, Anthropic updated its terms of service in 2026 to include global ownership restrictions: any company with more than 50% direct or indirect ownership by Chinese entities, regardless of where it is incorporated or operates, is prohibited from using Claude. The uniqueness of this incident lies in the fact that it was not Anthropic proactively pulling services from Goldman Sachs, but Goldman Sachs proactively consulting and tightening its own policies. This indicates that foreign financial institutions have begun incorporating AI tool geopolitical compliance into their standard risk management processes. The Goldman Sachs Hong Kong branch itself serves as a clearing partner, OTC trading counterparty, and compliance advisor for multiple major crypto institutions. If such tool restrictions become standard practice within the Hong Kong branches of foreign banks, the impact will extend beyond engineers in tech departments to business lines heavily reliant on AI assistance, such as crypto asset trading, structured product design, and client due diligence. It is currently unclear whether other foreign banks are taking similar measures. However, according to SCMP, as Anthropic strengthens identity verification (including KYC requirements), the scale of Hong Kong users accessing Claude via VPNs and underground channels is expanding—demonstrating that while demand exists, the compliant path has been blocked. This is not the first instance of an AI tool being restricted due to geopolitical factors, but its symbolic significance is exceptionally clear: when Anthropic's Claude becomes part of the core workflow of financial institutions, its service boundary is no longer just a product decision, but an extension of geopolitical reality. As the regulatory boundaries for AI between the US and China remain blurred, "which AI model you can use" is becoming a compliance issue, a geographical issue, and a political issue.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-24

Anthropic admits Claude has "gotten dumber": three engineering configuration errors, all subscription quotas reset as compensation

2026-04-24

Claude can now help you order Uber, play Spotify... connecting to eight daily apps, Anthropic is playing a big game

2026-04-23

OpenAI CEO Sam Altman slams Anthropic: Fear-mongering Claude Mythos is just to monopolize AI

2026-04-22

Bloomberg reveals Claude Mythos has been subject to unauthorized access! The hardest vulnerability for Anthropic to defend against will always be "people."

2026-04-21

Google assembles elite team to catch up with Anthropic, even its own engineers are secretly using Claude Code