An Inside Look at the Relay Market Powering Token Resellers and Fraud 透视驱动代币转售和欺诈的中继市场

TL;DR

• A black market for discounted LLM tokens exists, primarily in China, driven by resellers pooling API keys from various sources.
• Resellers obtain these keys through abusive practices such as exploiting free trials, proxying through unprotected support bots, and using stolen credit cards or chargeback attacks.
• The infrastructure relies on open-source proxy software like one-api and its fork new-api to load balance requests across pooled credentials.
• Buyers utilize these services to access cheap tokens, bypass geo-restrictions, and collect data for model distillation.
• The ecosystem incentivizes the exploitation of unprotected endpoints, prompting calls for stricter API key caps and limits from vendors.

Why It Matters

This development highlights a significant security and economic vulnerability in the current LLM API distribution model, where abuse of free tiers and weak authentication mechanisms creates a lucrative underground economy. For AI practitioners, it underscores the critical need for robust API management, strict usage caps, and secure endpoint configurations to prevent unauthorized access and financial loss.

Technical Details

• Proxy Infrastructure: Resellers utilize legitimate open-source tools like one-api and new-api to aggregate and distribute API credentials, effectively creating a relay network that masks the origin of the requests.
• Abuse Vectors: Key acquisition methods include abusing free trial periods, exploiting unprotected customer support bots, and utilizing fraudulent payment methods such as stolen credit cards and chargeback schemes.
• User Motivations: The demand side is driven by cost reduction, circumvention of geographic restrictions, and the collection of high-quality interaction data for potential model distillation efforts.
• Vendor Gap: Current API offerings often lack granular, hard-stop spending caps, allowing users to incur unexpected costs or be exploited without immediate service termination upon reaching a budget threshold.

Industry Insight

LLM providers must implement mandatory, strict spending caps and real-time anomaly detection to mitigate the risk of API key abuse and fraud. Developers should audit their public-facing applications for unprotected endpoints, as these serve as low-hanging fruit for the relay market, and consider implementing rate limiting and enhanced authentication measures.

TL;DR

• Matt Lenhard 深入调查了围绕 LLM 代币转售和欺诈形成的地下市场，主要集中在中国地区。
• 转售者通过滥用免费试用、代理未受保护的支持机器人或使用被盗信用卡等手段，以折扣价提供 API 访问权限。
• 该市场使用的代理软件（如 one-api 和 new-api）本身是合法的开源负载均衡工具，但被恶意利用。
• 买家追求低价、规避地理限制，部分用于模型蒸馏；这加剧了开发者对应用被滥用导致高额账单的担忧。
• 文章呼吁 LLM 供应商必须改进 API 密钥的严格限额功能，以在达到设定金额阈值时自动停止服务。

为什么值得看

这篇文章揭示了 LLM 生态系统中一个隐蔽但危害巨大的灰色产业链，展示了合法基础设施如何被转化为欺诈工具。对于 AI 从业者和供应商而言，它提供了关于 API 安全漏洞、滥用模式及防御策略的重要警示，有助于优化产品设计和风险控制机制。

技术解析

• 欺诈手段：转售者主要通过三种方式获取廉价令牌：滥用免费试用账户、利用未受保护的客户支持机器人接口进行代理转发，以及使用被盗信用卡或发起拒付攻击。
• 代理软件架构：核心工具为开源项目 one-api 及其活跃分支 new-api。这些软件设计初衷是合法的 API 网关，用于在多个 API 凭证池之间负载均衡请求，但在黑产中被用作聚合非法来源令牌的中间件。
• 市场需求驱动：终端用户购买此类服务的动机包括降低 Token 成本、绕过地域访问限制，以及收集数据用于后续的小模型蒸馏训练。
• 安全风险外溢：该市场的存在形成了一个激励生态系统，促使攻击者不断寻找新的未受保护端点进行利用，增加了公共 LLM 应用暴露面的风险。

行业启示

• API 计费与风控升级：LLM 供应商应强制或默认提供更细粒度的 API 密钥限额功能，支持按金额或请求量设置硬性上限，并在触发时立即阻断服务，而非仅依赖事后通知。
• 开发者安全意识强化：应用开发者需更加谨慎地公开 LLM 驱动的接口，实施严格的速率限制、身份验证和异常流量检测，防止自身应用成为黑产代理池的一部分。
• 开源中间件的双刃剑效应：开源 API 网关工具虽具价值，但缺乏内置的反滥用机制可能使其成为黑产温床；供应商和社区需加强对此类工具的安全配置指导和监控能力。