CVE-2026-27494: n8n: security flaw enables exploitation

n8n is an open source workflow automation platform. Prior to versions 2.10.1, 2.9.3, and 1.123.22, an authenticated user with permission to create or modify workflows could use the Python Code node to escape the sandbox. The sandbox did not sufficiently restrict access to certain built-in Python objects, allowing an attacker to exfiltrate file contents or achieve RCE. On instances using internal Task Runners (default runner mode), this could result in full compromise of the n8n host. On instances using external Task Runners, the attacker might gain access to or impact other task executed on the Task Runner. Task Runners must be enabled using `N8N_RUNNERS_ENABLED=true`. The issue has been fixed in n8n versions 2.10.1, 2.9.3, and 1.123.22. Users should upgrade to this version or later to remediate the vulnerability. If upgrading is not immediately possible, administrators should consider the following temporary mitigations. Limit workflow creation and editing permissions to fully trusted users only., and/or disable the Code node by adding `n8n-nodes-base.code` to the `NODES_EXCLUDE` environment variable. These workarounds do not fully remediate the risk and should only be used as short-term mitigation measures.

An attacker with a low-privilege n8n account (e.g., a contractor, compromised SaaS user, or insider) creates a new workflow containing a Python Code node. They craft Python leveraging unrestricted access to built-in objects — using introspection chains like `''.__class__.__mro__[1].__subclasses__()` to locate `subprocess.Popen` or `os.system` outside the sandbox boundary. On the default internal Task Runner configuration, this executes directly on the n8n host. The attacker reads `/proc/1/environ` and local `.env` files to harvest OpenAI API keys, vector database credentials, and cloud provider tokens. With RCE, they install a reverse shell, establish persistence via a cron job, and pivot to connected AI services — accessing the organization's RAG corpus, model fine-tuning datasets, and LLM inference infrastructure. The entire AI stack is now under adversary control.