Anthropic Tightens Blockade, Cracking Down on “Workaround” Access
According to a July 3rd report by the Financial Times, AI startup Anthropic is actively taking measures to close loopholes that allow mainland Chinese companies to access its AI models through various workarounds. Anthropic’s terms of service explicitly prohibit Chinese companies and their foreign-controlled entities from using its models, with enforcement that is among the strictest of its US peers. This includes requiring rigorous identity verification and banning payments with Chinese bank cards.
To enforce this policy, Anthropic continuously updates its detection and response tools using technical means. For instance, the company has previously used its AI model, Claude Code, to analyze user information like computer time zones to determine if a user is physically located in mainland China. Anthropic has publicly stated: “We explicitly prohibit access to or facilitation of access to Claude in unsupported regions, including China, and will identify and block non-compliant accounts through our evolving detection systems and collaboration with partners.”
Workaround Access by Tech Giants Was an “Open Secret”
Despite the strict ban, the demand from Chinese engineers for advanced AI models like Claude remains strong. Previously, accessing Claude through workarounds had become a common practice for some Chinese tech companies.
One primary method was using overseas subsidiaries and cloud services. Sources familiar with the matter revealed that Ant Group had provided corporate Claude accounts to employees through its Singapore-based entity via the company’s internal network. Another, more decentralized approach, was seen at ByteDance, where employees were allowed to expense personal Claude subscriptions and access the service using tools like VPNs. Furthermore, it became an “open secret” in the industry for cloud service providers like Microsoft Azure to sell API access to Chinese companies with overseas entities.
In addition to these corporate-level methods, “relay” services also exist in the market, forwarding requests from Chinese users to overseas-registered Claude accounts. However, large companies have generally been cautious about such services due to data security risks, such as prompts being stored or resold.
Alibaba Proactively Cuts Ties, Citing “Backdoor” Security Risks
On the same day Anthropic tightened its blockade, Alibaba Group made an internal decision to proactively “cut ties.” According to internal sources, Alibaba has added Anthropic’s products to its high-risk software list and required all employees to uninstall them by July 10.
Alibaba’s core reason is security risk. It is reported that a recent report pointed out a security vulnerability in Anthropic’s code model, Claude Code, suggesting it could have a “backdoor.” After an internal comprehensive assessment, Alibaba decided to ban the product to mitigate potential risks. As an alternative, Alibaba is recommending that employees use its in-house AI coding assistant, Qoder.
US-China AI Ecosystems Decouple Faster, Domestic Alternatives Prioritized
Anthropic’s blockade and Alibaba’s ban, two events occurring on the same day, have collectively elevated the use of frontier AI models from a simple matter of technical efficiency to a complex issue of corporate governance, data security, and geopolitical compliance.
This series of developments highlights the accelerating trend of decoupling between the US and Chinese AI ecosystems. Although top American models still hold reference value for technological R&D, the combination of difficulty in obtaining accounts, restrictions on cross-border data flow, and potential security risks is forcing Chinese companies to shift their focus more towards in-house tools and domestic large models. This incident will undoubtedly raise the priority of domestic AI solutions as replacements within Chinese companies, accelerating their technological iteration and application.
