[keorn]

First you need to understand what is a fault in this case. If you make use of PoW the chain is picked on the basis of total difficulty, this means that if any validator (already admitted by a centralised party) is hacked, then they can very quickly produce a chain which gets accepted by all others as canonical (with higher difficulty). This reverses any existing transactions and requires significant extra-protocol action to mitigate.

With PoA there is no such ability, since the canonicality is not abusable. Furthermore addition and removal of nodes can be decentralised and determined according to any rules (since validators can be specified in a contract). This makes it possible to include validators based on something like majority voting.

[decentralised]

I understand that if one node is compromised, it can then be used to generate malicious txs and the attacker can try to add them to a block it generates, but why would you say the other validators accept such block(s) or transactions in the first place?

It seems to me that a successful attack on a private blockchain requires the attacker gaining access to both a "pre-approved" node and having the hashrate majority needed to ensure its blocks are generated faster than the rest of the network combined.

Some background: https://coinjournal.net/vitalik-buterin-on-misconceptions-in... and https://www.reddit.com/r/ethereum/comments/48m9n6/vitalik_bu...