Yes, I noticed ambition last week when Claude found my local-only projecta .md file with server credentials and stuff and decided to attempt root access over SSH. Even though root access is disabled and I use a non-root sudo user, it still tried to act as root. No instruction to do that was given.
We were casually developing the project in VS Code, then pushing to GitHub, with the server pulling from GitHub.
We started with:
Claude does the coding, user check, user pushes to GitHub, then user pulls on the server.
Then:
2. Claude does the coding, realizes I use GitHub Desktop for one-click push, and decides to push directly to GitHub from VS Code.
And then:
3. Claude does the coding, pushes to GitHub, and tries to use root SSH to pull from GitHub, which got our IP blocked. Twice.
No instruction was given to do that.
Obviously, Claude learned the steps I repeat after each code edit and decided to execute them autonomously.
After that (two IP bans), it apologized and promised it would never attempt root SSH access on the server again.
So yes, AI takeover is already in process.
You can now enable Claude to use your computer to complete tasks.
It opens your apps, navigates your browser, fills in spreadsheetsβanything you'd do sitting at your desk.
Research preview in Claude Cowork and Claude Code, macOS only.