PraisonAI versions prior to 4.5.121 contain a critical OS command injection vulnerability (CVE-2026-40088) due to unsafe use of subprocess.run() with shell=True on user-controlled input. Multiple input vectors exist, including YAML workflow definitions, agent configuration files, recipe steps, and LLM-generated tool calls. This allows attackers to inject arbitrary shell commands via shell metacharacters, potentially leading to full system compromise. A fix is available in version 4.5.121.