AI proxy services are exposing users to significant financial, data, and privacy risks, according to recent findings. These intermediary platforms, which have emerged as solutions for developers in mainland China facing network, payment, and verification barriers, often offer AI services like Claude, GPT, and Gemini at unrealistically low prices. However, users risk losing their deposits if upstream accounts are shut down, as these services often operate without refunds or customer support. Research by the CISPA team highlights that some proxy services dilute model capabilities, with performance dropping significantly compared to official APIs. For instance, Gemini 2.5 Flash's performance on the MedQA benchmark plummeted from 83.82% to 37% when using certain shadow APIs. Additionally, these services may route requests to cheaper models while misrepresenting the model used, leading to inaccurate evaluations and compromised user experiences. Privacy concerns are also paramount, as relay servers can access and potentially misuse sensitive data, including business plans and personal information. Users are advised to avoid transmitting sensitive materials through these intermediaries, as they may be used for unauthorized training or resold. Warning signs include services offering prices far below official rates, lacking company transparency, and requiring large pre-deposits.