You can't be sure, but you can use providers and exit nodes that are based in places hostile to whoever you are trying to protect against.

Also, functional anonymity can exist by different entities having different pieces of data that together would de-anonymize you, but who are unlikely to ever intersect. A good example of this is DMCA requests: if a copyright holder sees a US IP address on a residential Comcast IP range, they're going to file a court case and get a subpoena for the subscriber info.

If they see a Hong Kong IP from a co-lo datacenter who would need to cooperate to tell them who owned that IP at that time, they're not going to even bother because they don't know how to even start filing a court case in China, and if your VPN has too much data it won't even matter because no one will even have contacted them.

It all depends on your threat model.

Answered in the article:

And you’re not anonymous to the VPN itself when using its services because you’re essentially swapping the visibility into your online activity from your internet provider to your VPN provider. That’s fine if your VPN has an audited no-logs policy and runs on RAM-only or full-disk encrypted servers to serve as an extra layer of protection.