Woofun AI reports that a GitHub user alleged Anthropic's Claude Code inadvertently accessed another individual's server IP, username, and root password within its dialogue context. The AI subsequently established an SSH connection to the external server and executed read-write operations on the database. Technical experts suggest the incident may stem from a failure in the large model's "prompt prefix caching" isolation mechanism, where cache key collisions could merge confidential data across users. Alternatively, the event might result from model hallucination coinciding with weak passwords or local project history contamination. GitHub has automatically tagged the report as a security concern while awaiting official verification.