Relays Are Masking Chinese Access to Frontier AI Models in the US

Refract AI Intelligence Digest

BLUF

Chinese actors are using a massive network of relay servers to evade sanctions and steal US frontier AI capabilities.

NEWS

Intelligence reports indicate more than 80,000 AI relay servers are actively masking the identities of users in China accessing cutting-edge large language models hosted in the US. This infrastructure facilitates unauthorized access, with analysts suspecting the primary goal is to clone proprietary models rather than legitimate use.

Why I Care

This undermines US AI leadership and national security by allowing adversarial nations to rapidly close the technology gap without investing in original research. It poses significant risks to intellectual property protection and could lead to the proliferation of dual-use AI capabilities.

Next Steps

AI providers should implement stricter geolocation verification and anomaly detection for API traffic immediately. Policymakers need to update export controls to cover relay infrastructure, while security teams must monitor for unusual access patterns originating from known proxy networks by Q4 2026.

More than 80,000 AI relay servers are helping users in China mask their identities while they access cutting-edge large language models (LLMs), probably to clone them.
Back to Blog Listing

Source: Dark Reading ·

This digest was generated by Refract AI Collective to help the public sector security community stay informed.