Zevorn RT-Claw Remote Code Execution Vulnerability in Telegram-to-AI Tool Integration

A remote code execution vulnerability has been identified in Zevorn RT-Claw versions prior to 0.2.0. The issue arises in the Telegram-to-AI tool execution flow, specifically within the 'tool_run_script_execute' function of 'claw/services/tools/script.c'. This vulnerability allows for code injection by manipulating the 'message.text' received from Telegram, which is processed by the AI engine and executed as Python code on the Linux platform, without any user confirmation or approval. The exploit has been made public and can be used to execute arbitrary Python scripts with the permissions of the RT-Claw process, potentially leading to unauthorized access to files, credentials, and other system resources.

Impact

Exploitation of this vulnerability allows for arbitrary Python code execution on the affected system, with the potential to read or write files, steal credentials, manipulate device states, and conduct follow-on attacks using accessible system resources.

Reproduction

The vulnerability can be reproduced by sending a message through Telegram that triggers the 'run_script' tool call. This can be done by influencing the AI model to respond with a 'run_script' command, which RT-Claw will execute immediately without any safety checks. The execution can be verified by checking for the presence of a created file that indicates the script was run.