Quote from: prof7bit on June 06, 2013, 10:39:07 AM

Quote from: donut on June 05, 2013, 07:02:35 PM

To clarify: difficulty adjusts every 2016 blocks. There is no time limit on this. If 2016 blocks were mined in one minute - the difficulty will adjust such that the next 2016 will be mined roughly 10 minutes apart from each other.

but if they then switch off this monster miner again and hashing goes back to previous speed then difficulty will remain to be 20160 times too high, it will then take 201600 minutes to mine a block, this is 140 days per block or 2.3 years for 6 blocks. It will take 773 years until the next difficulty adjustment.

I think a few people in here need to read this about how the target is drawn up:

Quote

For reasons of stability and low latency in transactions, the network tries to produce one block every 10 minutes. Every 2016 blocks (which should take two weeks if this goal is kept perfectly), every Bitcoin client compares the actual time it took to generate these blocks with the two week goal and modifies the target by the percentage difference. This makes the proof-of-work problem more or less difficult. A single retarget never changes the target by more than a factor of 4 either way to prevent large changes in difficulty.

Note that this shoudl be combined with this statement for full understanding:

Quote

The difficulty is adjusted every 2016 blocks based on the time it took to find the previous 2016 blocks. At the desired rate of one block each 10 minutes, 2016 blocks would take exactly two weeks to find. If the previous 2016 blocks took more than two weeks to find, the difficulty is reduced. If they took less than two weeks, the difficulty is increased. The change in difficulty is in proportion to the amount of time over or under two weeks the previous 2016 blocks took to find.

Taken from
https://en.bitcoin.it/wiki/Difficulty and
https://en.bitcoin.it/wiki/Target

Edit: This means, that no single entity can mess with the difficulty more than a single factor of 4 at a time. This would give people(devs) time to notice that something was up and code/propose a fork.
The attacker could, slowly (at each difficulty increase), bring their hardware online, increasing in hashing power multiples of 4. This would increase the difficulty by the maximum 4x every 3.5 days, up to the maximum achieveable level of their hardware hashrate. It would then be possible for them to perform the aforementioned switch-off, to leave the blocks unmined for (potentially) a long time, which again might require a hard fork of some kind to rectify.