[1] Cost estimate for a rollback

Let's assume a scenario where a rollback could be considered much more likely to occur:
1. Exchange "Octopus" is being hacked, 100.000 BTC are stolen.
2. Octopus' admins notice the hack "right away" after a measly 20 confirmations on the blockchain.
3. Octopus' CEO gets on the phone and calls his CEO friends at some of the major Mining Pools.

Now it's obvious that Octopus could create a double spend of the transaction(s) that stole the coins and submit those transactions to the miners.
The miners could decide to specifically mine a new longest chain starting with the double spend, thereby undoing the theft of the coins.
Obviously, Octopus would have to pay the miners for this rollback.

The direct cost of such a double-spend-rollback can be estimated as
- block rewards of all rollbacked blocks
- fees of all roll-backed transactions

Also, to help acceptance of the rollback, it would probably be wise to give all block rewards that were in the "original chain" to their respective "owners", as well as any transaction fees. That way, if I were a miner who received a few coins in the original chain, I would also receive the same amount in the new chain, making it less likely for me to oppose the rollback. [2]

All in all, I estimate the ground cost for a rollback at somewhere on the order of 30 BTC per confirmation.
Of course, the rollbackers would also expect to get their share, and probably they wouldn't accept a pittance.
So, as a lower bound for the real cost of a rollback, I estimate something on the order of at the very least a hundred BTC per confirmation.

Now that doesn't include any potential "side costs" for the rollback, e.g. the implications it would have an the resulting effect on the exchange rate of Bitcoin. That's much harder to estimate, but let's just say that there is probably some point where the miners (or rather: a majority of them) would consider it more profitable to commit themselves to the rollback than to stay true to the original chain.
Maybe a thousand BTC per confirmation will do the trick.

---

[2] Gentlemen's Agreement

In order to lower the resistance of other participants of Bitcoin against such a rollback, it would probably be wise to agree on some ground rules:
1. revert as few transactions as possible
2. re-instantiate all other transactions of the rollbacked chain
3. grant all block rewards & fees in the new chain to their respective recipient addresses in the rollbacked chain
Of course, this cannot be enforced, but would probably be self-enforcing, since it would most likely be cheaper to avoid resistance rather than fight it.

The point is:
1. they already have that power
2. centralization is not a result of this

The reason why this is not centralization is quite simple: it's costly.
The basic premise of the Bitcoin blockchain
still holds true.

The security of a transaction is not determined by the fact that it cannot be rolled back, but rather by the fact that it is costly to do so.


That can be mitigated by Octopus' CEO.
He could simply "pay" his miner friends by creating a double spend and some follow-up transactions with huge fees.
I.e. he first creates a double spend of the stolen coins to address A.
2. Tx A -> B 100 BTC fee
3. Tx B -> C 100 BTC fee
etc. pp.


Yup, that adds to the cost, making it highly unlikely to be able to pull off a rollback, if you don't have overwhelming support.
I.e. a simple 51% majority most likely won't do the trick, you'd rather go for something on the order of at least 75%, preferably more.

And yes, in the case of a "long rollback", i.e. a rollback after more than just a few dozen confirmations, the Bitcoin network would effectively be stalled for quite some time.

Such rollback had been done in the history of Bitcoin and is a useful tool in drastic circumstances.
But consensus of simple majority of miners could not be considered as the agreement of the bitcoin community.
There should be some sort of voting mechanism in between miner and holders to reach a consensus as these two groups are the ones most affected.

Why not just have a withdraw queue on your exchange? Max btc withdraw withing 1 minute = .5 btc, Hacker might get 1-100 btc then before reports come in.
Why not just have a max account withdraw queue on your exchange? Only 2-3 withdraws at a time to be processed manually and a way notice they are from the same i.p. or when mass accounts are trying to withdraw at the same time.

Why not, just not give another man your bitcoins to begin with, it defeats the whole purpose of being your own bank.

A rollback is like a kid spilling milk and expecting to go back in time to clean up the milk and making everyone else go back in time to, rather than grabbing a cloth and cleaning it up. We all cannot go back in time because of ones mistake.

Roll backs should be out of the question unless it was like that time they forked the coin and made 8.2 billion bitcoins or something.

Those stolen bitcoins are still gonna get back to the market and the entire community shouldn't worry about that.
People gave to Binance their money. Binance is an official and totally legal exchange so if they lost someone else's money then they have to give it back. It is only their problem and I bet that top crypto exchange can afford itself to repay this money. It is only a problem of Binance, not the entire cummunity.
The other thing is that mining community is not that smart and they don't really care about the others. It mainly consists of greedy (and pretty often retarded) asians and for obvious reason, I don't trust them.