CVE-2026-5971 Details

A critical eval injection vulnerability has been identified in FoundationAgents MetaGPT versions through 0.8.1. The issue resides in the XML Handler component, specifically within the ActionNode.xml_fill function of the file metagpt/actions/action_node.py. This vulnerability arises because the function uses the unsafe eval() method to process strings from LLM responses into Python objects, particularly for list and dict field types. An attacker who can manipulate the LLM's output, either through prompt injection or by exploiting a compromised model, could inject arbitrary Python code that gets executed on the server, leading to remote code execution.

Users are advised to update to a version of MetaGPT that addresses this vulnerability. If no such version is available, consider using the QEMU microVM sandboxed code execution feature, which can be integrated into MetaGPT as an alternative to the current host-process execution.