It seems likely (and perhaps obvious) that:

- people will try to abuse it

- abuse will keep it from supporting naive trust of semantic information published by untrusted third parties

But we're also already roughly in this scenario, and it seems like it might be easier to model and spot/discard abuse of semantic information.