According to a report from Dongcha Beating AI, security researcher Shou Chaofan stated that he recently purchased approximately 6TB of Fable model invocation data from a leading Chinese model transfer station. This data contained sensitive credentials such as SSH keys, VPN configurations, Alibaba Cloud keys, and GitLab tokens. He claimed that the keys in this batch of data were sufficient to allow him access to the servers or internal systems of 19 top Chinese companies, as well as 7 government-related agencies in China and the Commonwealth of Independent States (CIS), including Huawei, Xiaomi, NIO, and MiniMax. The model transfer station acts as an intermediary between users and models like Claude, meaning that requests and responses pass through it, allowing it to see the complete plaintext. If developers include SSH keys, API keys, VPN configurations, etc., in the Agent context, and the transfer station saves or even sells these records, the company's system keys could also be leaked. This is not the first time Shou has warned about the risks of transfer stations. In a paper he participated in, released in April, 428 LLM transfer stations were tested, revealing that 9 actively injected malicious code, 17 called AWS using intentionally placed test keys by researchers, and 1 directly transferred ETH from a test wallet. Shou Chaofan is a co-founder of blockchain security company Fuzzland and has long been involved in vulnerability and supply chain security research. At the end of March, he was the first to discover that the source map in the Claude Code 2.1.88 release package inadvertently exposed approximately 500,000 lines of TypeScript source code.