You've just watched a platform nuke an account you followed for the third time. No explanation, no appeal, no human on the other end of the process. You close the tab and start wondering if the whole architecture is broken.
It might be. And some people are building something structurally different, though "better" depends entirely on what you're optimizing for.
Decentralized social networks don't have a single moderation team reading your posts. They use a patchwork of overlapping, sometimes conflicting authority spread across thousands of independently run servers. Every server sets its own rules. The network itself is just the plumbing connecting them.
The server is the unit of government
Take Mastodon, the most widely used piece of the Fediverse. When you sign up, you're not joining "Mastodon" the way you join Facebook. You're joining a specific server, say mastodon.social or infosec.exchange, each run by a different person or organization with their own moderation policies, their own staff, and their own tolerance for what flies. That server talks to thousands of others through a shared protocol called ActivityPub. Your posts travel across the network. The rules governing them don't.
When a server admin decides they've had enough of a particular type of content, they have two tools. They can silence a remote server (posts from it stop appearing in their users' feeds but aren't deleted network-wide) or defederate from it entirely, severing the connection. Think of it like a country imposing a trade embargo: the other country still exists, its citizens still talk to each other, they just can't reach yours anymore.
This is what actually enforces norms. Not an algorithm. Not a policy team in a glass building.
Collective social pressure between server operators.
What a real moderation call looks like
Imagine two people, Priya and Marcus, who both joined Mastodon on the same day. Priya picked a large general-purpose server. Marcus joined a smaller, tightly run server focused on open-source software, with three volunteer moderators and a written code of conduct that fits on one page.
A week in, a spam account on a third server starts flooding the network with low-grade harassment. Priya's large server is slow to act; it has thousands of reports in the queue. Marcus's server defederates from the offending server within six hours because one moderator knows the admin personally and makes a call. Marcus never sees the harassment. Priya sees it for three more days.
Same network. Wildly different experience. That's not a bug in the architecture, exactly. It's the architecture.
What people consistently misread about this model
The common assumption is that decentralization means no moderation, a libertarian free-for-all where anything goes. That's wrong in both directions, and the people who repeat it haven't looked closely enough. Some federated servers are dramatically stricter than any mainstream platform, banning categories of speech that Twitter or Bluesky would permit without blinking. Others are permissive in ways that would get a post removed in seconds on Reddit.
The real tradeoff isn't between moderation and no moderation. It's between consistent, scalable, top-down moderation (one company, one set of rules, one billion users) and inconsistent, local, community-driven moderation that's highly responsive in small spaces and patchy at scale.
Trust-and-safety researchers keep circling back to a stubborn problem with the decentralized model: coordinated harassment campaigns, state-sponsored disinformation, and viral radicalization pipelines are genuinely harder to address when there's no central actor who can pull a network-wide lever. The Fediverse had visible struggles with this as it grew. Individual server admins, often unpaid volunteers, found themselves making consequential calls about novel threat patterns with no legal team, no threat intelligence feed, and no budget. Governance by exhausted enthusiast has real limits.
Bluesky takes a different approach with what it calls the "composable moderation" model: a central protocol layer, but labeling and filtering handled by independent services called labelers. You subscribe to the moderation feeds you trust, stack them, and your timeline reflects that stack. Modular rather than federated. Different tradeoffs again.
So before you pick a server, do you actually check its moderation history? An active, transparent block list and a responsive admin are worth more than a slick interface. Most people skip this step and then complain about the neighborhood.
The deeper point is that content moderation was never really a technical problem. It's a governance problem. Centralized platforms made it look technical by hiding the governance inside a single company's decisions, which is a neat trick if you can get away with it. Decentralized networks put the governance back where it always was: in human communities, with all the inconsistency and occasional brilliance that implies.