Anthropic closes loopholes allowing Chinese access to Claude AI: Report
Anthropic accused Alibaba-affiliated entities of carrying out a massive-scale distillation campaign in mid June
Anthropic is set to plug the loopholes in Claude that enabled Chinese companies to bypass the usage restrictions and access its AI models unauthorized.
According to the sources privy to the matter who told the Financial Times that the various Chinese companies including Ant Financial has had unauthorized access to Anthropic’s Claude Code “through workarounds.”
Different Chinese companies deployed different methods to access Claude, For instance, ByteDance implemented a reimbursement scheme enabling engineers to use personal subscriptions for Claude via VPNs. Some other companies used Microsoft’s Azure cloud services by using foreign subsidiaries.
Last week, the tech giant accused Alibaba-affiliated entities of carrying out a massive-scale distillation campaign in mid June, alleging these operators used around 25,000 fake accounts to perform 29 million exchanges.
The reported infringements occurred despite Anthropic’s strictest bans on usage in China as it requires user verification and block payments from Chinese banks.
In the wake of growing breaches, Anthropic has moved to close the loopholes that are compromisable and stepped up efforts to restrict Chinese use through loopholes, as reported by sources familiar with the matter.
The company also explicitly prohibits access from unsupported regions and targets "transfer station" services that relay requests from China to overseas accounts.
Anthropic also announced to constantly update their tools to identify unauthorized access and take enforcement actions against them.
“We explicitly prohibit accessing or facilitating access to Claude in unsupported regions, including China. Anthropic is the only frontier AI company that restricts sales to PRC-controlled companies, including subsidiaries incorporated outside China,” the company said.
“It enforces this policy through continuous, evolving detection systems, working alongside our partners to identify and ban accounts that violate our policies.”
-
OpenAI autonomous agents probed Hugging Face for vulnerabilities weeks before major breach
-
Apple plans AI servers with M8 ultra chips and Nvidia technology: What to know
-
Altman says AI accidents are coming: Here's his plan
-
Can we really prevent an AI extinction event?
-
Google quietly adds selfie video sign-in for account recovery
-
Billionaire Mark Cuban says your feed may be bigger AI threat
-
iOS 27 lets you hide your location without alerting anyone
-
EU steps up AI safety push with frontier labs talks