Arbitrary Code Execution Vulnerability in Langflow's Agentic Assistant Feature

CVE-2026-33873

What is CVE-2026-33873?

Langflow, a platform for building AI-driven agents, contains a vulnerability in its Agentic Assistant feature that allows untrusted input to be executed as server-side Python code. This flaw arises during the validation phase of generated components, where it improperly reaches dynamic execution sinks. Attackers with access to this feature can manipulate model outputs to generate and execute arbitrary code on the server, posing significant security risks to applications using this tool. The issue has been addressed and resolved in version 1.9.0.

References