Executive brief
zevorn rt-claw is an AI assistant framework. A security vulnerability in its networking tool allows an attacker to force the assistant to make unauthorized web requests to internal systems or local services. This could result in the exposure of sensitive internal data, such as administrative endpoints or cloud metadata, which are normally protected from external access.
Technical details
A Server-Side Request Forgery (SSRF) vulnerability exists in the `http_request` AI tool of zevorn rt-claw up to version 0.2.0. The vulnerability is located in the `claw_net_get` and `claw_net_post` functions within `claw/services/tools/net.c`. The application fails to validate or filter the `url` argument provided by the AI model (which can be influenced by user chat input) before passing it to the Linux networking backend. On Linux systems, this URL is passed directly to libcurl via `CURLOPT_URL` in `osal/linux/claw_net_linux.c`. An attacker can exploit this to perform arbitrary HTTP requests to loopback addresses (127.0.0.1), private RFC1918 networks, or cloud metadata services. Because the tool returns the response body back to the AI engine for chat output, this is a non-blind SSRF that allows for direct data exfiltration. As of the advisory date, no patch has been released.
Affected products
- zevorn rt-claw up to 0.2.0
Timeline
- 2026-06-12: disclosed: Issue reported to the project maintainers via GitHub issue #134
- 2026-07-18: advisory: CVE-2026-16125 published by VulDB/NVD