I'm not sure exactly, but I suspect it goes something like this:

An algorithm they are using are finding a match in some category(ies) for you that other people have already searched or it might be an AI trained model. Like there are people you know or close to your network, that searched for the same thing and they clicked the specific link.

Now use a curl in a vpn in a different network and you get totally different results.

This was my first thought as well, but it turns out not to be the case. Using curl in a VPN turns out roughly the same results. I suppose it's possible that enough people have found out and sought out the relevant news that the engine is making the association though.

When you do an exact match search by putting the name in quotes, does it bring anything? It should bring up content that contains the name and even highlight it.

Yes, and these articles don't show up. There are no strings that point directly to his name in the AMP pages that report on it.

In this case could it be matching against a cached, earlier publication where the name was (accidentally) published before being edited out in the version you are seeing now?

I suppose it's possible, but I haven't seen any traces of it yet. I'll look through archive.org, but it's all regional news in Canada.

What do you mean? It sort of negates the implicit discretion that one would hope for by redacting the name of someone involved. Seems kind of akin to Facebook revealing the sexual identity of someone who hasn't revealed that to certain people.