| Title | vanna.ai vanna 2.0.2 Remote command execution |
|---|
| Description |
Because **user input is controllable**, and LLM output is also affected by prompt injection, attackers can construct malicious prompts to induce the model to insert arbitrary Python statements (e.g., importing `os`/`subprocess` and executing system commands) into the "Plotly code". If this code is executed by `exec()`, it could result in **arbitrary command execution (RCE)** on the host machine.
|
|---|
| Source | ⚠️ https://github.com/Ka7arotto/cve/blob/main/vanna-rce.md |
|---|
| User | Goku (UID 80486) |
|---|
| Submission | 03/06/2026 12:46 (1 month ago) |
|---|
| Moderation | 03/20/2026 15:28 (14 days later) |
|---|
| Status | Accepted |
|---|
| VulDB entry | 352077 [vanna-ai vanna up to 2.0.2 /src/vanna/legacy exec injection] |
|---|
| Points | 20 |
|---|