Simon Willison: An Inside Look at the Relay Market Powering Token Resellers and Fraud

Main idea

An extensive underground marketplace — primarily in China — exists where resellers acquire cheap or stolen LLM API access (OpenAI, Anthropic, etc.) and resell it at a profit. The infrastructure is built on open-source proxy software one-api and new-api.

Context

Willison highlights Matt Lenhard's investigation from July 26, 2026 (vectoral.com). The relay market exists due to pricing and geographic restrictions — resellers arbitrage between access cost and sale price. Customers include businesses seeking cheaper access, those bypassing geoblocks, and those collecting data for model distillation.

Why it matters

Willison emphasizes a critical security implication: the existence of this ecosystem means there is now an entire network of actors with financial motivation to find new unprotected LLM endpoints. He recommends LLM vendors implement strict spending caps on API keys.

Details / arguments

• Relay market operates via one-api/new-api open-source proxy software
• Token sources: free trials, unprotected support bots, stolen cards, chargebacks
• Customers: cheaper access, geoblocking bypass, model distillation data
• Willison: now more cautious about publicly exposing LLM endpoints
• Solution: strict spending caps on API keys (feature most vendors have but rarely surface prominently)