A lot of this reasoning is from before Datachains. As it stands now, if you gain control of a section, you own the network. You can change any mutable data in the section (which would include stealing the safecoin), and you can probably expand your attack past your section.
You won’t be able to decrypt any data, which is a small victory, I guess.
Any number (65%, 80%, etc) was probably based on designs before data chains and node age. I don’t think anyone has done the analysis of the current design. And with age, that analysis is going to be exceedingly complex. I have not even seen any discussion on what the actual size of a section will be.
There may be defenses for some or all of these attacks, or it may be (hopefully will be) extremely hard to gain control of a section. There are also clever moves an attacker can do to increase their chances. For instance, if you own 30% of two adjacent sections, DDOS your neighbors to force a section merge. Complex systems have large attack surfaces.
We will see how it all shakes out.
Per the current documentation (can’t find the original source at the moment Data Chains documentation summary):
Security of the network depends on each section behaving correctly, which is ensured by the network distributing nodes throughout the network automatically, such that an attacker cannot choose where in the network any nodes he runs will join, and a brute force attack would require running a significant proportion of the nodes in the network to have any real chance of success… The broken section recovery mechanism could be used as a weapon allowing a section controlled by an attacker to take over more of the network, even up to the point of taking over the entire network if the attacker had enough resources to handle network load; this is a known limitation, but necessarily any mechanism to recover from a state where no consensus is possible must reduce security. This does not necessarily mean the system is less secure than a network using a less strong form of consensus, and is only an exploit if an attacker can control one section in the first place.