Anthropic Tightens Enforcement Against Chinese Access to Claude Amid US-China AI Decoupling

Anthropic is actively closing loopholes that allowed Chinese technology firms, including Ant Financial and ByteDance, to access its advanced Claude AI models through overseas subsidiaries, VPNs, cloud providers, and corporate networks, according to a Financial Times report citing people familiar with the matter.

The moves come days after the US Commerce Department lifted short-lived export controls on Anthropic's Fable and Mythos models, which had been imposed over national security concerns related to potential jailbreaks exposing sensitive capabilities.

Chinese entities reportedly used workarounds such as routing access via Singapore-linked entities (in Ant Financial's case, partly owned by Alibaba) and expense reimbursements for personal subscriptions accessed through VPNs. These methods breached Anthropic's terms of service prohibiting use by Chinese companies or Beijing-controlled entities but did not violate laws.

Anthropic has targeted "transfer station" services that relay requests from mainland China through overseas accounts. The company stated it prohibits access from unsupported regions like China and continuously updates detection systems.

Separately, Anthropic has accused Alibaba of running the largest known "adversarial distillation" campaign, using thousands of fraudulent accounts to extract capabilities from Claude, prompting a letter to US senators.

These actions highlight a deepening pattern of controlled access to frontier AI models, extending beyond government export rules to private enforcement of usage restrictions.