CVE-2026-54769

Langroid is a framework for building large-language-model-powered applications. Versions prior to 0.65.2 are vulnerable to a critical Sandbox Escape leading to Remote Code Execution (RCE) in its TableChatAgent and VectorStore capabilities. When these agents evaluate LLM-generated tool messages with full_eval=True, they attempt to sandbox the execution by explicitly setting locals to an empty dictionary {} inside Python's eval() function. However, this relies on an incomplete understanding of Python's execution model. Because builtins is not explicitly scrubbed from the globals dictionary mapping, Python implicitly injects all built-ins during execution, granting full access to functions like import('os').system(). Since TableChatAgent.pandas_eval() executes external LLM outputs natively, this bypass permits any attacker providing prompt payload to achieve unauthenticated RCE on the host system. Version 0.65.2 patches the issue.