Anthropic has blocked access to Claude in China as part of its compliance posture around export controls and geopolitical risk, but a new WIRED investigation shows that determined users in the country are routinely getting through anyway. VPNs, foreign SIM cards, and third-party API wrappers are among the tools people use to reach the model, raising uncomfortable questions about how effective these restrictions actually are.
The Gap Between Policy and Practice
Anthropic's decision to cut off users in China is not unique among US AI companies. The restrictions reflect both regulatory caution and a broader strategic calculation about where American AI should and should not operate. That tension has been explored in depth by analysts watching the US-China AI race, and the Claude blackout has already created openings for Chinese competitors like Zhipu GLM-5.2 to fill the void. But the WIRED report suggests the blackout is far from airtight on the demand side.
Key Facts
- Anthropic enforces access restrictions in China through IP-based geolocation detection.
- Users report successfully bypassing blocks using commercial VPN services and foreign phone numbers.
- Third-party platforms that wrap the Claude API also provide indirect access.
- Demand for Claude inside China remains high despite the restrictions.
- Experts say geolocation enforcement is structurally difficult to make airtight.
The workarounds are not sophisticated by technical standards. Most users rely on commercial VPN services that route traffic through servers in the United States, Japan, or Europe. Others obtain foreign phone numbers or payment methods to complete account registration. A smaller group accesses Claude indirectly through third-party products built on the Claude API, where the geolocation check happens at a different layer. Anthropic has previously outlined scenarios in which US AI leadership depends heavily on controlling who gets access to frontier models, making the persistence of these workarounds a genuine policy concern.
"Geolocation is a speed bump, not a wall. Anyone with moderate technical knowledge and a reason to try will get through."Security researcher quoted by WIRED
What Anthropic Can Realistically Do
Tightening enforcement is possible but comes with tradeoffs. More aggressive IP blocking can catch traffic from known VPN exit nodes, but that approach also risks cutting off legitimate users in other countries who happen to route through the same servers. Requiring stricter identity verification at signup could reduce access from China, though it creates friction for users globally and raises its own privacy considerations. Anthropic has not publicly commented on the WIRED findings or indicated whether it plans to change its enforcement approach.
The company's public position has focused on responsible deployment and safety rather than on the mechanics of geographic enforcement. Its leadership, including Dario and Daniela Amodei, who were recently named to TIME's 100 Most Influential People, have spoken broadly about the importance of AI development going well for everyone, a framing that sits awkwardly alongside policies designed to keep entire populations from using the product at all.
None of this is to say the restrictions are pointless. They do create meaningful friction, and most casual users in China will not bother with the workarounds. The policy also signals intent, which matters for regulatory relationships in Washington. But the WIRED investigation is a useful reminder that digital borders are porous by nature, and that a company's formal access policy and its real-world access footprint are rarely the same thing. As Claude grows in capability and visibility, that gap is likely to attract more scrutiny, both from regulators watching for circumvention and from users who simply want access to a tool they cannot easily get.
