Bribing outside of the system (i.e. in real life) could be a tremendous coordination and logistical challenge, especially given the amounts needed to make these attacks profitable. Smart contracts can make all of that trivially easy.

It’s like... the people most interested in trustless collaboration are people who are untrustworthy. A trustless settlement layer seems fine, but any real human application is going to have an element of trust and doesn’t belong alongside settlement.