This collection of networks offers no end to end encryption. Anyone with administrator access to an Instance can read anything that travels through that Instance’s infrastructure – including direct messages. The level of risk correlates with the number of cross-Instance interactions between users. If users from different Instances communicate, an attacker need only compel one Instance to reveal the direct messages between all of the interacting accounts. The centralised equivalents – Twitter, Tumblr, etc – can cloak their users through governance and resources. In a peer-to-peer network without encryption, there’s no structure, no agreed-upon governance, and absolutely no protection. Compromising or compelling an Instance or its staff means that all of network traffic is laid bare to its assailant.
I’d love to have a discussion on this (now fairly old) article which IMO has yet to provoke the kind of much-needed action on this topic that we, as a community of cypherpunks, are capable of.
Well, the malicious actors can setup their own instances as well and exploit the inherent trust between the participants by design. P2P sold as security property in the scenario where participants are unknown and multiple in numbers is misconception. It does not square well with basic security mindfulness, and shouldn't be taken as improvement in that regard.
I think that federation and all this stuff is not about improving security, it is a form of grassroots communication based on certain principles. If you need security, you use other tools, and treat these things as public, hostile spaces.