CVE-2026-109440 Details

A vulnerability in LangAlpha, an open-source AI agent for financial research, allows remote code execution through indirect prompt injection. This issue is present in LangAlpha versions prior to 2026.09.07. The vulnerability arises because the agent can execute shell commands or Python code without human approval, by embedding instructions in web content that the agent retrieves. Exploitation of this vulnerability enables attackers to execute arbitrary code in the user's workspace sandbox and steal sensitive vault secrets from a specific internal file.

It is recommended to gate the Bash and ExecuteCode tools behind human-in-the-loop approval, especially when the agent is handling untrusted web content. Additionally, do not rely solely on the natural-language security prompt to protect vault secrets; instead, implement explicit controls to manage sandbox egress and keep sensitive data out of reach of unapproved code execution.