Executive brief
IBM Langflow OSS, a tool used to design and deploy AI workflows, contains a critical security flaw that allows unauthorized individuals to take full control of the system. By exploiting a combination of a weak default login setting and a flaw in how the software runs Python code, an attacker can execute commands on the host server without needing a password. This could lead to the theft of sensitive AI provider keys (like OpenAI or Anthropic), access to private databases, and complete compromise of the underlying infrastructure.
Technical details
A critical vulnerability exists in IBM Langflow OSS (versions 1.0.0-1.9.3) involving two primary issues. First, the default configuration (LANGFLOW_AUTO_LOGIN=true) allows unauthenticated users to obtain a superuser JWT via the /api/v1/auto_login endpoint. Second, the PythonREPLComponent fails to properly isolate execution; while it attempts to restrict globals, it does not explicitly set the '__builtins__' key to an empty dictionary. In CPython, the exec() function automatically restores full access to the builtins module if it is missing from the globals dictionary, allowing attackers to use functions like __import__, open, and eval. An unauthenticated remote attacker can leverage these flaws to execute arbitrary OS commands with the privileges of the backend process (often root in Docker environments) and exfiltrate sensitive environment variables or flow definitions. The issue is resolved in version 1.9.4.
Affected products
- IBM Langflow OSS 1.0.0 through 1.9.3
Timeline
- 2026-06-21: advisory: Initial publication of IBM security bulletin
- 2026-06-22: disclosed: CVE published to NVD dataset
- 2026-06-21: patched: Remediation available in version 1.9.4