# Node-RED, the 10.0 that is already being exploited in installations without login

> CVE-2026-58115 allows code execution without authentication in Node-RED. Full exploitation has been under way since October 30 and it affects industrial equipment.
> By José Maldonado · Cybersecurity · 2026-11-05

Node-RED was installed so that an operator could connect a machine to a spreadsheet without writing a line of code, and that same ease is what now turns it into an open door. The tool's HTTP interface does not require authentication by default, so anyone who reaches it can create a new flow and execute code with the process's permissions. No password is needed, nor a session, nor a memory flaw to exploit.

The problem is registered as [CVE-2026-58115](https://cyberstrike.io/cve/CVE-2026-58115/) and Siemens, which distributes the tool within its industrial system, scores it 10.0, the maximum rating. VulnCheck observed the first full exploit on October 30 and added it to its catalog of exploited vulnerabilities, according to the patch roundup published by [heise's Patchwork newsletter](https://www.heise.de/) on November 4. CISA had not added it to its list when that report was published.

The exploitation observed is not a laboratory experiment. Attackers deploy a Linux bot and webshells, which is what gets installed when the goal is not to steal data but to stay inside. The flaw affects Node-RED installations in any sector, and more uncomfortably the industrial equipment that ships with it from the factory and that nobody updates because nobody knows it has Node-RED inside.

## A 10.0 attack, a security nightmare

In practice, the Node-RED editor is a full web application with access to the operating system behind it, and the only barrier between the two is a login screen that has to be enabled by hand.

The vulnerability is a lack of authentication in the HTTP interface. The flow editor is designed to be opened in a browser and to drag blocks around, and in that default configuration it does not ask for credentials from anyone who enters. An attacker with network access can open the editor, create a flow with a node that executes system commands and wait for the flow to trigger.

The 10.0 rating is not an exaggeration. Such a score is reserved for flaws that are exploited remotely, without authentication, without user interaction and with total impact on confidentiality, integrity and availability. All four apply here: there are no credentials to steal, there is no victim who has to click and the resulting code runs with the permissions of whoever started the service.

Comparing it with other flaws from this month helps to place it. The two 9.8s in [LangFlow in its AI agents](https://observatorioblockchain.com/en/cybersecurity/langflow-cve-2026-93680-dos-fallos-criticos-agentes-ia/) required a specific configuration; this one requires nothing more than the default installation.

## Linux bot and webshells in installations without login

What has been seen being installed is a Linux bot and webshells, two pieces that usually go together. The bot turns the machine into part of a command-and-control network and gives the attacker a door that survives reboots; the webshell gives them convenient HTTP access without opening new ports, something valuable in networks where only 80 and 443 go out to the internet.

A webshell on a server that also runs Node-RED is especially troublesome because the legitimate service already runs with broad permissions. The attacker does not need to escalate privileges: they inherit the ones the process had to talk to connected devices, which in an industrial environment include PLCs, sensors and protocol gateways.

The detail that sets this incident apart from any ordinary web compromise is the context. A compromised Node-RED in an office is just another server; a compromised Node-RED in a plant is a write path into the physical process, and that changes the order of priorities in any response.

## An attack capable of touching the PLC

Siemens [warned](https://cert-portal.siemens.com/productcert/html/ssa-834709.html) in August that devices running Industrial OS earlier than version 4.3.4.1 shipped with Node-RED inside, including the SIMATIC IoT2050 Advanced, an industrial gateway used to carry plant data to management systems. Version 4.3.4.1 or later fixes the flaw in that environment, and the company's advisory is the only one that offers a clear version number for the industrial estate.

Outside Siemens things get more complicated, because Node-RED is installed by hand anywhere: a Raspberry Pi in a cabinet, a container in the cloud, an integration server someone set up to connect the labeling machine to the ERP. None of those cases has a manufacturer publishing a bulletin, and they all share the same default configuration.

The official recommendation for those installations is concrete and can be applied today: enable \`adminAuth\`, which requires authentication before opening the editor, and do not expose the service directly to the internet. Neither of the two closes the vulnerability, but both eliminate the entry path being used.

## Three measures that work today without patching

The first is editor authentication. \`adminAuth\` is configured in the settings file and supports users with a password or with tokens, and from that point on the interface is no longer anonymous. It is the highest-impact and cheapest measure: a configuration change and a service restart.

The second is to close off the network path. If Node-RED only needs to talk to four devices, it does not need to be reachable from the entire corporate network. A firewall in front of the editor port, or a segmentation rule, reduces the number of people and machines that can try without touching the software.

The third is to check whether the editor has been published unintentionally. Internet scans find Node-RED instances with the editor open with surprising ease, and they are often test setups from two years ago that nobody shut down. Searching for your own before someone else finds them is an afternoon well spent.

## How to detect an already compromised installation

A compromise of this kind leaves traces at three different layers, and they should be reviewed in order, because each one explains what the previous one cannot: first the application, then the system and finally the network.

The most direct sign is flows that nobody wrote. The editor stores flows in a configuration file and in its internal storage, and any flow with a command execution node that does not correspond to a documented process deserves an explanation. In an installation with dozens of flows, comparing against the previous week's backup settles the question in minutes.

Then you have to look at the system, not the application. A process that starts from a temporary path, a new scheduled task, an SSH key added to a service user or a periodic outbound connection to an unknown address are the bot's fingerprints. The case of [PwnEscape, the flaw that lets a guest attack the host](https://observatorioblockchain.com/en/cybersecurity/pwnescape-cve-2026-99999-escape-invitado-anfitrion/), left the same trace on the affected devices.

The third place is Node-RED's own log. New flows leave deployment entries with a date and user, and on an installation without authentication that user appears blank. A deployment with no author at three in the morning is exactly what you should be looking for.

## The new boundary between OT and IT

The tool that was born so that an operator would not have to depend on the IT department has ended up being the point where that department loses control. Node-RED is useful precisely because anyone can put together an integration in an afternoon, and that autonomy is what leaves installations outside the inventory, unpatched and with no identifiable owner.

The [AI harness that helped breach 200 organizations](https://observatorioblockchain.com/en/cybersecurity/artex-vapor-panda-ia-pentesting-200-organizaciones/) showed the other side of the same problem: when integration software has no owner, it also has no one to defend it. The asset list of any plant should already include the glue tools, not just the equipment that produces.

While that happens, this week's task fits in three lines. Locate the organization's Node-RED instances, enable \`adminAuth\` on all of them and check whether any has flows that no one recognizes. The rest, including the debate over who should have warned earlier, can wait until Monday.
