Investigation uncovers underground market reselling discounted LLM tokens in China
Resellers pool API keys via open-source proxies to undercut pricing, exploiting free trials, unprotected bots, and stolen payment methods.
2 sources · cross-referenced
- An investigation describes a growing market in China where resellers offer discounted LLM tokens by pooling API keys from multiple sources.
- Resellers use open-source proxy software like one-api and new-api to load-balance requests across pooled credentials.
- Buyers seek lower costs, evade geo-restrictions, or collect data for model distillation.
- Operators exploit free trials, unprotected support bots, stolen credit cards, and chargeback attacks to reduce costs.
An investigation by Matt Lenhard documents a market primarily active in China where resellers aggregate and resell access to large language model (LLM) tokens at discounted prices. The market operates by pooling API keys from various sources to create a proxy service that undercuts standard API pricing.
The proxy infrastructure relies on open-source software, notably one-api and its actively developed fork new-api, which are designed to load-balance requests across multiple API credentials. These tools are legitimate API management products repurposed for resale operations.
Buyers of these discounted tokens include entities seeking cost savings, those aiming to bypass geo-restrictions, and actors collecting data for model distillation purposes.
Operators of these resale networks employ several tactics to reduce costs, including abusing free trial offers, routing traffic through unprotected support bots, and using stolen credit cards or chargeback attacks to fund API usage. This ecosystem creates a profit motive for continuously discovering and exploiting unprotected endpoints.
Commentary accompanying the report emphasizes the need for LLM vendors to implement stricter controls on API keys, such as hard spending caps, to prevent abuse and unexpected billing shocks for developers.
- Jul 26, 2026 · TechCrunch — AI
Monday.com cites AI in 20% workforce reduction, joining 20+ tech firms with AI-linked layoffs in 2026
Trust74 - Jul 25, 2026 · MIT Technology Review — AI
AstraZeneca details AI’s expanding role in biologics drug discovery and R&D acceleration
Trust78 - Jul 25, 2026 · Simon Willison — everything
PyPI begins rejecting new file uploads to releases older than 14 days to mitigate supply-chain risk
Trust80