CVE-2026-61446 describes a remote code execution vulnerability in MervinPraison's PraisonAI before version 1.6.78. The plugin manager uses importlib's spec_from_file_location() and exec_module() to load Python plugins from project-level and user-home .praisonai/plugins/ directories without code signing, integrity checks, or sandboxing. If an attacker can write a malicious Python file to these directories—via path traversal, supply chain compromise, or other means—they can achieve arbitrary code execution during plugin initialization. The vulnerability has a CVSS 4.0 score of 8.6, indicating high impact with low attack complexity and no required privileges or user interaction.

Successful exploitation allows an attacker to execute arbitrary Python code remotely on the affected system without requiring privileges or user interaction. This can lead to full system compromise, data theft, or further malicious activity. The vulnerability arises from the lack of code signing and integrity verification in the plugin loading process.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, restrict write access to the .praisonai/plugins/ directories to trusted users only and monitor for unauthorized file additions. Avoid using untrusted plugins or dependencies that could write to these directories.