unless i am misundestanding. not sure how this computer prevents secrets from my gmail leaking. thats the worst case.

If you put your gmail credentials into a VM that an AI Agent dealing with untrusted prompts has access to they should be treated as leaked and be disabled immediately.

However, if you don't put your administrative credentials inside of the VM and treat it as an unsafe environment you can safely give it minimal permissions to access specific things that it needs and using that access it can perform complex tasks.