Goldman Sachs blocks Hong Kong staff from Anthropic — UK banks watch closely
TL;DR: Goldman Sachs has removed access to Anthropic’s Claude for its bankers in Hong Kong, with sources telling Reuters the move followed a strict reading of the bank’s contract with Anthropic. ChatGPT and Gemini remain available on the same internal platform. The decision marks the first major Western bank carving out a regional Anthropic ban — a precedent UK financial services compliance teams now have on file.
The Financial Times broke the story on Tuesday. Reuters confirmed that Goldman staff in the territory previously interacted with Claude through an internal AI platform but lost access in recent weeks. Goldman declined to comment; Anthropic told the FT its Claude models had never been officially supported in Hong Kong.
What’s actually driving the carve-out
Two factors are converging. First, Anthropic’s latest model — Mythos — has drawn fresh scrutiny from banks and financial regulators after demonstrating capabilities that AISI used in its evaluation work earlier this month. The Hong Kong Monetary Authority told Reuters last week some banks were assessing additional mitigation measures in response to evolving AI cybersecurity threats. Second, US-China tensions over AI access are spilling into a grey zone: Hong Kong has historically sat outside the China bans US firms apply to ChatGPT and Claude, but that gap is narrowing. Anthropic’s own market disclosures do not list Hong Kong as supported.
Goldman’s CIO Marco Argenti said in February the bank was working with Anthropic to develop AI-powered agents to automate internal functions — making this carve-out more strategically interesting than a simple vendor reshuffle. The bank is keeping the broader Anthropic relationship intact while drawing a hard line at one geography.
Looking forward
For UK banks, the read-across is direct. Most run vendor risk assessments that already segment AI tooling by user location. Goldman’s move shifts the precedent: a contract-based carve-out by jurisdiction is now the live template. Compliance teams at UK SME-facing banks and asset managers should expect questions about regional access controls in the next round of internal AI governance reviews. Whether this proves to be the first cut of many — Singapore, mainland exposure, third-party data residency — depends on how the HKMA’s mitigation guidance lands and whether other AI vendors follow Anthropic’s posture.