A majority of hashpower can add softforks. They don't even have to publish software that are compatible with them, but they could.
For all we know-- there are some of these active now, invisible to us. -- though probably not.
The way carlton described it up thread-- it's not obvious why it could even be a problem. If it was just some new transaction feature and it was useful and not bothering anyone else... then it might well be not a problem. But it can take other forms where it clearly is a problem.
This has been raised as a weakness in Bitcoin's history-- and some have looked to solve it. Usually the context it discusses is softforks that block a particular user (e.g. Bad Guy Bob) or style of transactions (e.g. Lightning payments). Sometimes people have been concerned with fluffyforks--
ones where the rule will be bypassed if a violating chain gets ahead.
But it appears to be fundamentally hard to solve because a miner imposing a soft-fork is indistinguishable from an unmodified miner operating behind a network filter that hides certain transactions/blocks from it. You could make sure they weren't getting filtered if there was some consensus about their inputs ... yo dawg I hurd u like blockchains.
There is obviously the nuclear option: If a majority hashpower is disruptive enough the users of Bitcoin will take some extreme action like changing to a new POW ... But not only would that path be dangerous and disruptive (making it a less potent threat), it's unlikely to work if the attack is only harming a few users ("so sad Wikileaks can't move their funds, but not my problem").
So the open questions are: Are their less extreme defences than the nuclear option, is there a way to
amplify attacks so that any amount of attacking will trigger the political will to fight back, can malicious softforks be stopped?
There has been advancement in a number of these dimensions.
Detection: now that the network is operating against the weight limit most of the time there is little reason to have valid transactions rejected (except for forward compatibility reasons)--- transactions really should be being accepted in a fee-rate sensible order, and if they're not thats clear and detectable. So if it started happening at any scale, we could probably see it.
Amplification: Coinjoins and sendmanys link together the spends of multiple users, you can't block one input in a transaction without blocking them all. More researchy right now, but
non-interactive aggregate signatures have the property of building transaction bundles that cannot be unlinked. Both result in disruption amplification and additional fee income loss for miners that censor.
Indistinguishably: Taproot should make different kinds of transactions much more indistinguishable (and an older technology, the
coinswap transform can have similar indistinguishability benefits without taproot), and if usage types can't be distinguished they can't be targeted. Similarly, privacy improving practices (avoiding reuse, coinjoins, swaps, anti-network-monitoring, p2p encryption) can get in the way of censoring based on who's transacting.
Anonymous mining: The system's existing incentives discourage leaving out any economical transactions -- after all, you'll miss out on their fees. But external pressures might make you act in ways that aren't rational within the system. Tech like compact blocks, satellite block transmission, and not bloating the hell out of the maximum block size preserves the at least the threat that a lot of mining could go dark.
Mining distribution/security: These attacks assume a majority hashrate is colluding against user interests. That's easier if mining is more centrally controlled. Betterhash/stratum2 make some steps in the direction of improving that. There is probably a lot more to do to further decentralize mining control-- and this area has probably the least progress and some of the most importance for this issue.
In any case, -- I hope this is some food for thought.
