Remote Code Execution in PraisonAI via LLM-Generated Python Code

Executive Summary

CVE-2026-61447 is a critical vulnerability in the PraisonAI CodeAgent component, specifically in versions prior to 1.6.78. The vulnerability exists because the _execute_python() method executes Python code generated by a large language model (LLM) without proper validation or sandboxing.

This method runs the code in a subprocess that inherits the full environment of the parent process, including all sensitive secrets such as API keys and credentials. Despite a sandbox setting being enabled, it is effectively ignored.

Attackers can exploit this by injecting malicious prompts that influence the LLM output, allowing them to execute arbitrary code remotely and exfiltrate all environment secrets.

Useful 0 Not Useful 0

Impact Analysis

This vulnerability can have severe impacts including full credential theft and arbitrary code execution on the host system.

An attacker who successfully exploits this flaw can remotely execute malicious code, potentially taking full control of the affected system.

The exposure of sensitive environment secrets such as API keys and credentials can lead to further compromise of connected systems and data.

Useful 0 Not Useful 0

Mitigation Strategies

To mitigate this vulnerability, immediately upgrade the PraisonAI CodeAgent component to version 1.6.78 or later, where the issue has been addressed.

Avoid using versions of the praisonaiagents package prior to 1.6.78, as they execute LLM-generated Python code without proper sandboxing, AST validation, or import restrictions, allowing remote code execution and credential theft.

Additionally, review and restrict any external inputs that influence LLM output to reduce the risk of prompt injection attacks.

Useful 0 Not Useful 0