Executive brief
zevorn rt-claw is an AI assistant framework that integrates Telegram with AI models. A security flaw allows remote Telegram users to execute arbitrary Python code on the host system without any user approval. This could lead to full system compromise, data theft, or unauthorized access to the host environment where the assistant is running.
Technical details
A code injection vulnerability exists in the Telegram-to-AI Tool Execution Flow of zevorn rt-claw. The root cause is the lack of an approval mechanism or 'deny-by-default' policy in the tool dispatch loop within `claw/services/tools/script.c`. When a remote Telegram user sends a message that causes the AI model to return a `run_script` tool call, the `tool_run_script_execute` function immediately executes the provided Python code via the Linux platform backend. An attacker can exploit this by influencing the AI model to trigger the `run_script` tool, resulting in arbitrary code execution with the privileges of the rt-claw process. As of the advisory date, the project has not responded to the issue report.
Affected products
- zevorn rt-claw up to 0.2.0 and unreleased upstream commits
Timeline
- 2026-07-18: disclosed: Vulnerability reported via GitHub issue #138
- 2026-07-19: advisory: CVE-2026-16204 published