AI Controlling your browser - what could possibly go wrong
I just found out about a new feature (as of Dec 11, 2025) that Chrome has added to Chrome DevTools. This is done with an MCP Server running in your browser. The AI agent running on your local machine can then connect to the browser and do things the local AI agent asks it to do.
This is a blog explaining this feature:
https://developer.chrome.com/blog/chrome-devtools-mcp-debug-your-browser-session</br>
It starts with “Many of our users have been asking for this” - lol
This Chrome Developers YouTube channel video announces the feature:
https://youtu.be/2rOeZ98AOb8?si=ZwLz_Y8RYtvxxO57&t=46
I’m not advocating this feature. Letting AI run your browser could be extreamly damaging if it “halucinates” what it thinks you wanted it to do. As Steve Gibson would say, https://www.grc.com/sn/sn-822.htm#:~:text=what%20could%20possibly%20go%20wrong
Chrome’s blog (above) says :
There are safeguards. Remote debugging is disabled by default, and must be explicitly enabled at chrome://inspect/#remote-debugging, and every new debugging session requires user approval. Also while a session is active, Chrome shows the banner: “Chrome is being controlled by automated test software.”
So this is good to know.
NOTE: This feature was first released in Chrome 144 (Jan 7, 2026) and as of now in Chrome v146 (March 10, 2026) this feature is not enabled by default.
To see the currernt setting use this URL:
chrome://inspect/#remote-debugging
Developers wishing to run this in a development environment in a virtual machine with fake accounts can enable this for experimentation purposes.