Ah, my recollection was that only a fixed (large) set of parties could challenge the withdraw with the fraud proof, so it was also a security assumption, but maybe that was BitVM v1? The nice thing ab...
Section 8.1 of BitVM2 says that challenges are open; that since this offers a griefing vectors, challengers should post collateral; and that since that disincentivizes, they add on a crowdfunding element with sighash flags. I suspect (but definitely don't know) that what this means in practice is, challenging could be made to work if the costs aren't exorbitant, but the costs are pretty exorbitant in the basic BitVM(2) design - section 8.5 talks about 4MB txs. Possibly if something like Glock+Argo mac or BABE or similar actually ends up working, the whole thing becomes more practical. But with very chunky assert transactions it's pretty problematic. But very vague uncertain comments, here.