They're trying to bolt an unpredictable language engine onto an OS whose security model is based on total predictability.

It's basically a rootkit that you consent to installing. No audit logs will help after the agent has already exfiltrated your data because a malicious prompt was hidden in a white pixel on an image it was analyzing

Lol. Windows already is malware. Anyone who's running Windows is fundamentally fine with not being in control of what's happening on the computer they're using.

Upvoted anyway. People should be aware of the ancillary risks of pointing Redmond's loaded gun at their own head, even if they accept the inherent risks to doing such.

Will this be like the agentic AI which truncates entire tables full of data in your database and then lies about it?

You: "You installed something with malware while I explicitly said you should not do it."

Copilot: "I didn't install anything at all."

...

It's almost as if shoving AI into any available nook or cranny, regardless of whether it's even remotely useful, is a bad idea.