Claude Code is accused of the risk of context isolation failure, and user credentials may be misused

A user reported on GitHub that when using Anthropic's Claude Code, a server IP, username, and root plaintext password that did not belong to them appeared in the conversation context, causing the AI to automatically connect to other servers and modify the database. Community technicians analyzed that the root cause of the problem lies in the failure of the prompt prefix cache isolation mechanism in the large model, which leads to collisions between cache keys of different users. Currently, GitHub has tagged the issue as' area: security 'and all parties are waiting for Anthropic's verification conclusion.