PraisonAI versions prior to 4.5.115 are vulnerable to remote code execution due to unsafe YAML deserialization in the AgentService.loadAgentFromFile method. The vulnerability arises from using the js-yaml library's load function without restricting dangerous tags, allowing malicious YAML files to execute arbitrary JavaScript code on the server. An attacker able to upload a crafted agent definition file can trigger this execution, potentially compromising the server fully. A fix is available in version 4.5.115 that uses a safe YAML schema to prevent code execution.