Skip to content
NewA calmer, faster ZeroHour
AI safety & securityhighExploited in the wild

Stolen AI credentials feed growing LLM proxy economy

CSO Online ()4 min read7 · 5 reads
AI summary · grok-4.7Summaries by Vani AI

Team Cymru estimates more than 80,000 proxy servers hide stolen AI credentials and enable model distillation.

Team Cymru identified transfer-station proxies running Claude Relay Service and sub2api, later estimating more than 80,000 such servers. The relays authenticate customers locally but call frontier APIs with pools of stolen keys and subscriptions, with clusters tied to China and Hong Kong while many gateways sit on US VPS hosts. Palo Alto Networks, Okta, and Gambit Security separately documented token-jacking, including hundreds of valid Anthropic, OpenAI, Gemini, and other credentials harvested by infostealers and phishing. NSA, CISA, and FBI recently accused six China-based AI companies of industrial-scale distillation of US models through similar stations.

Key facts
  • Researchers first found 10,867 CRS and sub2api servers; the estimate now exceeds 80,000.
  • More than 4,000 China and Hong Kong IPs hit 304 US-hosted relays, moving about 14TB.
  • Seventeen Anthropic relays uploaded about 81GB and received 1.4GB, a 58-to-1 ratio.
  • Stolen keys included Anthropic, OpenAI, Gemini, Groq, OpenRouter, xAI, and AWS credentials.
  • NSA, CISA, and FBI accused six China-based firms of industrial-scale model distillation.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

From the original article

4 min read at csoonline.com

Cyber threat groups have increasingly targeted enterprise AI assets, such as credentials, cloud environments, and research, as a means for operationalizing their own use of AI. Now, another sophisticated means for obfuscating illegitimate use of AI resources is coming more clearly to light.

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.