> But what happens if a miner interprets a Segwit transaction as "pay to anybody" and creates a block that violates the "additional validation"
They don't because segwit (or taproot, for that matter) script is flagged using input-space that the old software knows is "from the future" so that it knows it doesn't know how to validate it. As a result, they won't relay it, won't mine it, and won't display it in their wallets until confirmed. But if someone else puts it in a block they'll accept it.
If someone goes and lobotomizes their own software so that it'll mine stuff it knows it doesn't understand, then they could produce an invalid block. But anyone that is upgraded just ignores their block like it never happened.
This is why the activation of such things is normally triggered by a super-majority of hashpower being upgraded: It's for the benefit of parties that haven't upgraded so that even if there are some bad-data-maniacs out there pulling expensive stunts, their invalid blocks are quickly left behind and even non-upgraded nodes won't see many confirmations before the bad block is removed. (And upgraded nodes won't see any at all, of course.)
In the case of taproot 90% of the hashrate has to signal support during any one of several two week signaling periods to trigger activation. If that doesn't happen, people will figure out why and try again (potentially with different activation criteria).