It'd be tricky to avoid false positives

You could check Wikipedia instead. If it says he's dead on Wikipedia for longer than a few hours/days, then he's probably dead.

Yeah, there are a lot of ways it could fail. You could probably build a pretty reasonable signal by looking at a lot of trends on Google News, Twitter, Wikipedia, etc. If you were building this in far enough in advance you could also test it by looking for high probability events besides your own death or demise.

What I like about it is that you could build a fairly hermetic system that, once you set it up, is never directly touched by you and just accesses public, high-volume websites. If you can set it up without leaving any fingerprints, that seems like a big plus.