That's not what this is intended to protect against. If your local machine is compromised, you are screwed, and 2FA no matter what the form isn't going to help you. At a minimum, for that, you need out-of-band confirmation of critical actions (e.g., a message to a different pre-registered device to confirm financial transactions before they are executed.)

If there's malware on your computer, it could just as easily keylog your password and one-time code the next time you try to log in to google, silently drop those packets on the floor (so you think there was just a connection issue), and then use your credentials to get access to your account.

If there's malware on your computer running as you, with access to things like USB devices, it becomes significantly harder (if not impossible) to do anything security critical on it.

At the very minimum, this requires physically interacting with the security key (by touching it), which prevents software from silently using the key to authenticate in the background. Then again, if your entire machine is compromised you basically have a MITM attack so close to your side of the channel that it is virtually impossible to do anything about it.

Yes, I think the protocol protects against MITM and replay attacks. http://fidoalliance.org/specs/fido-security-ref-v1.0-rd-2014...