Soumettre #773890: https://bagofwords.com/ bagofwords <=0.0.297 Remote command execution

The `bagofwords` application, intended as an AI agent for data analysis, contains a critical Remote Code Execution (RCE) vulnerability. The application allows users to ask questions about their data, which prompts an LLM to generate Python code (specifically a `generate_df` function) that is subsequently executed on the server.

Due to the lack of input sanitization and insufficient sandboxing of the execution environment, an attacker can use a prompt injection attack to manipulate the LLM's output. By instructing the model to insert malicious Python commands (e.g., using `__import__('os').system(...)`) into the generated function, arbitrary system commands are executed on the host server when the application processes the query. This flaw enables attackers to bypass intended restrictions and gain unauthorized control over the server environment.

• La source: ⚠️ https://github.com/bagofwords1/bagofwords/issues/60#https://github.com/Ka7arotto/cve/blob/main/bagofwords-rce.md
• Utilisateur: Goku (UID 80486)
• Soumission: 06/03/2026 12:09 (il y a 3 mois)
• Modérer: 20/03/2026 13:38 (14 days later)
• Statut: Accepté
• Entrée VulDB: 352065 [bagofwords1 bagofwords jusqu’à 0.0.297 code_execution.py generate_df élévation de privilèges]
• Points: 20