A critical remote code execution vulnerability has been identified in Xinference versions 2.5.0 and earlier. The vulnerability arises from the improper handling of attacker-influenced Llama3 tool-call outputs passed to the Python eval() function in multiple components, including xinference/model/llm/tool_parsers/llama3_tool_parser.py and xinference/model/llm/utils.py. Exploitation occurs when malicious requests are sent to the /v1/chat/completions endpoint with a tools field, allowing attackers to execute arbitrary Python code on the server. This vulnerability carries a CVSS score of 10.0 and poses a critical risk to systems utilizing the affected software. All users of Xinference should immediately apply the latest security patches and upgrade to a version beyond 2.5.0 to mitigate this risk. No authentication is required to exploit this vulnerability, making it particularly dangerous. Organizations should conduct network monitoring for unusual activity and ensure proper input validation mechanisms are in place until a patch is applied.