Soumettre #856869: zevorn rt-claw unreleased post-v0.2.0 (36d128f) Server-Side Request Forgery (CWE-918)information

Titrezevorn rt-claw unreleased post-v0.2.0 (36d128f) Server-Side Request Forgery (CWE-918)
Description# Technical Details A Server-Side Request Forgery vulnerability exists in the `http_request` AI tool in `claw/services/tools/net.c` of rt-claw. The application fails to validate destination URLs before performing outbound HTTP requests. When the AI model (or an attacker influencing the model) returns a `tool_use` block naming `http_request` with an attacker-controlled `url`, the tool passes that URL directly into `claw_net_get()` / `claw_net_post()`, which on Linux hands it to libcurl via `CURLOPT_URL` without loopback or private-network filtering. The response body is returned through the tool result, enabling non-blind SSRF with response exfiltration. # Vulnerable Code File: claw/services/tools/net.c (lines 447–463) Method: tool_http_request() Why: The function reads `url` from the attacker-influenced tool-call JSON and forwards it directly into `claw_net_get()` / `claw_net_post()` with no destination validation. The Linux backend in `osal/linux/claw_net_linux.c` (lines 189–235) passes the unvalidated URL to `curl_easy_setopt(curl, CURLOPT_URL, url)`. # Reproduction 1. Build rt-claw for Linux with `-Dosal=linux -Dtool_net=true` 2. Start a localhost canary HTTP server on `127.0.0.1` with a `/internal-only` endpoint 3. Start a mock LLM API that returns a `tool_use` response for `http_request` with `url="http://127.0.0.1:<port>/internal-only"` 4. Run the rtclaw binary and send input that triggers the AI chat flow 5. Observe the canary server receives a hit from 127.0.0.1 and the tool result body contains the canary token 6. Run control with no tool call — no localhost request occurs # Impact - Attacker can read localhost-only admin/debug endpoints, private RFC1918 services, and cloud metadata endpoints - Non-blind SSRF: response bodies are returned directly in the tool output channel - Enables internal network reconnaissance and data exfiltration from services reachable by the device
La source⚠️ https://github.com/zevorn/rt-claw/issues/134
Utilisateur
 Eric-y (UID 95889)
Soumission12/06/2026 11:37 (il y a 1 mois)
Modérer17/07/2026 18:08 (1 month later)
StatutAccepté
Entrée VulDB379837 [zevorn rt-claw jusqu’à 0.2.0 http_request net.c claw_net_get/claw_net_post url élévation de privilèges]
Points20

Interested in the pricing of exploits?

See the underground prices here!