Researchers: Malicious code can hide via AI chatbot Grok

Security firm Check Point Research has shown that malicious software (malware) on Windows can use AI assistants such as Grok from xAI and Microsoft Copilot to hide its contact with criminals. That is according to several security outlets. The researchers published the technique on February 18, 2026, and call it "AI in the Middle".

It is important to understand that this is a demonstration by researchers. The sources do not describe the method being found in widespread attacks. It also assumes that an attacker has already gotten malware onto the victim's computer.

How the method works

Malware normally needs to talk to its operators to get orders. That contact is called a command-and-control channel, often abbreviated C2. Usually, security programs can detect suspicious traffic to criminal servers.

In the experiment, the malware instead talks to an AI assistant. It sends a message asking the AI to fetch and summarize a web page that the attacker controls. Hidden commands sit on the page. The AI reads them and passes them on to the malware. This creates a channel where data can flow both ways.

On Windows, the malware can use a built-in browser component called WebView2. It can open the AI service in a hidden window so the user sees nothing.

Why it is hard to detect

The traffic goes to well-known and trusted AI services that many companies let through. It therefore looks like ordinary use.

At the same time, according to the researchers, the method requires neither an account nor an API key. An API key is a kind of password for the service. When there is none, the provider cannot simply close an account or revoke a key to stop the abuse. The researchers also say that encryption can fool the services' own security checks.

What does it mean for you?

For ordinary users, the advice is the same as before: keep Windows and your programs updated, avoid suspicious files and use an up-to-date security solution. The method does not change the fact that malware first has to get onto the computer.

For IT departments, the point is that AI services cannot automatically be considered safe in network traffic. They should consider monitoring which programs contact AI services, and why.