Artificial intelligence startup Anthropic is actively identifying and patching the workarounds used by Chinese entities to access its network.
Earlier, a report by Cryptopolitan showed that Anthropic believed Chinese tech firms have been accessing its Claude AI using indirect channels. Most of them have been relying on proxy cloud services and international subsidiaries to access Anthropic’s AI lineup.
While these methods do not necessarily violate U.S. or Chinese law, they breach Anthropic’s terms of service, which prohibit Chinese companies and their foreign affiliates from using its AI models.
In the last few months, the California-based AI research company has been imposing some of the harshest restrictions on China among U.S. AI firms, mandating identity checks and blocking all transactions with Chinese banks.
Anthropic’s coding technology has become highly sought after by Chinese engineering teams looking to replicate its AI capabilities.
To safeguard its models, the firm has long maintained one of the AI industry’s strictest access policies regarding China. Beyond blocking users based on geography, the company also verifies identities, restricts payments originating from Chinese financial institutions, and monitors unusual account activity.
Recently, the tech company accused Alibaba of bypassing its geographic restrictions. It claimed Alibaba deployed over 25,000 fraudulent accounts to log 28.8 million interactions with its Claude AI system.
Sources familiar with the matter have also claimed that Chinese corporation Ant Financial gave employees corporate Claude accounts that they accessed through the company’s internal network, tied to its Singapore-based entity. Moreover, according to five insiders at TikTok’s parent company, ByteDance, engineers bypass access issues by using VPNs and expensing personal Claude subscriptions.
This follows prior allegations by Anthropic against DeepSeek, Moonshot, and MiniMax for unauthorized use of Claude’s outputs. It documented 150,000 queries from DeepSeek, 3.4 million from Moonshot, and 13 million from MiniMax.
Some of these Chinese companies have been utilizing Microsoft Azure cloud services through overseas branches to connect with Claude, enabling developers in mainland China to access the AI via corporate networks. People familiar with the matter said Microsoft provided API access to Chinese firms with Singapore-based subsidiaries.
Nonetheless, a source familiar with the practice contended that the use of overseas entities to access Claude is well known and extends beyond a single provider. In response, Microsoft also said Anthropic monitors the service and enforces its terms and conditions with its assistance.
At the moment, Anthropic is reportedly working to shut down “transfer station” services that channel requests from mainland China through overseas Claude accounts before relaying the responses. However, big Chinese AI companies usually avoid using middleman servers because they fear operators will steal or sell their prompts to competing labs.
Nevertheless, it emphasized that it regularly refines its tools to detect and respond to new techniques Chinese users use to bypass restrictions. It added that it will keep taking action against accounts found to be in breach of its policies. Previously, it used Claude Code to determine a device’s time zone to verify whether a user was located in mainland China.
So far, Chinese authorities have banned companies from relying on overseas AI models for consumer-facing applications due to cross-border data regulations. The restrictions do not apply to internal R&D conducted by AI laboratories.
Aside from its latest crackdown efforts, Anthropic is looking to expand. The firm said Thursday that it has secured a lease for 113,000 square feet at Seattle’s Dexter Yard complex. The space represents nearly one-third of the campus at 700 Dexter Ave. N., located in South Lake Union’s tech hub. Before the new lease, the company occupied approximately 56,000 square feet at the Dexter Yard complex under a separate short-term arrangement.
While the lease terms remain undisclosed, CoStar, the commercial real estate analytics company, said the agreement marks the biggest office lease signed in Seattle this year.
Meanwhile, the firm brought Fable 5 back to global users this Wednesday after the U.S. government lifted the trade blocks that shut it down on June 12. It said the model will be available within users’ weekly usage limits through July 7, capped at 50%, before moving to a usage credit system. Mythos 5, on the other hand, will only be available to a set number of U.S. firms.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.